A method, system and electronic device for constructing virtual elastography imaging

Abstract

The application discloses a kind of virtual elastic imaging construction method, system and electronic equipment, the method includes: establishing the B mode feature-hardness distribution prior knowledge base of each type of lesion;Based on the mapping relationship in B mode feature-hardness distribution prior knowledge base, implicit mechanical characteristic extraction is carried out to B mode image to be processed, and the tissue mechanics potential representation of B mode image to be processed is obtained;According to tissue mechanics potential representation, the lesion related area of B mode image to be processed generates the elastic imaging information of B mode image to be processed;According to B mode image to be processed, the background area of B mode image to be processed generates the image background information of B mode image to be processed;Elastic imaging information and image background information are fused, and the virtual elastic imaging of B mode image to be processed is obtained, and virtual elastic imaging can accurately reflect the hardness difference and distribution characteristics of organization.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence processing technology, and specifically to a method, system, and electronic device for constructing virtual elastic imaging. Background Technology

[0002] Endoscopic ultrasound (EUS) is an important diagnostic and assessment tool for digestive system diseases, and it has significant clinical value in the evaluation of biliary and pancreatic diseases, submucosal tumors of the gastrointestinal tract, and peri-digestive lymph nodes. Traditional B-mode endoscopic ultrasound images can display the location, size, shape, boundaries, echo intensity, internal echo distribution, and relationship with surrounding tissues of lesions, making it the most commonly used basic image form in clinical endoscopic ultrasound examinations. Furthermore, endoscopic elastography can further reflect mechanical information such as tissue stiffness, tissue strain, and compressive compliance on top of B-mode images, providing a reference for lesion characterization, risk assessment, puncture target selection, and follow-up observation.

[0003] However, in actual clinical practice, some equipment / examination procedures do not use elasticity mode, operators do not routinely collect or save data, and previous data is incomplete. Furthermore, existing elastography technology usually relies on real-time elastography acquisition or obtains elastography images based on raw ultrasound echo signals, tissue deformation information before and after compression, displacement estimation, or strain calculation. As a result, not all cases can acquire or save high-quality elastography images in real time, causing a large number of existing B-mode images to be unable to further explore their potential tissue stiffness information, thus limiting the in-depth application value of endoscopic ultrasound in retrospective studies, remote consultations, and auxiliary diagnosis.

[0004] Therefore, existing technologies suffer from the problem of being unable to effectively assist in endoscopic ultrasound research due to the lack of elasticity imaging. Summary of the Invention

[0005] This invention provides a method, system, and electronic device for constructing virtual elastography, aiming to solve the problem that the lack of elastography in existing technologies makes it difficult to effectively assist in endoscopic ultrasound research.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for constructing virtual elastic imaging includes: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain a first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. For the same type of lesion, feature mapping is performed on the B-mode structural features and the relative hardness distribution features to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related region of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed; The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

[0007] Optionally, the acquisition of the sample data set includes: Filter the initial sample data pairs from the same case, the same examination procedure, the same lesion, or the same image scene; By removing outlier data pairs, incomplete data pairs, and artifact data pairs from the initial sample data pair set, a first sample data pair set is obtained. The first set of sample data pairs is subjected to standardization preprocessing and spatial registration to obtain the set of sample data pairs. The standardized preprocessing includes at least one of image cropping, size normalization, grayscale normalization, brightness and contrast correction, noise suppression, removal of irrelevant text or borders, image orientation unification, color space conversion, and image format conversion. The spatial registration includes at least one of deep learning, affine transformation, or non-rigid deformation.

[0008] Optionally, the first region set includes the lesion region, the adjacent reference tissue region, and the background tissue region; the step of dividing the B-mode image into regions to obtain the first region set of the B-mode image includes: Target recognition is performed on the B-mode image to obtain a lesion region mask, and the lesion region is determined based on the lesion region mask; The region in the B-mode image that is at a preset distance from the lesion region is defined as the adjacent reference tissue region; The remaining area in the B-mode image, excluding the lesion area and the adjacent reference tissue area, is defined as the background tissue area.

[0009] Optionally, the B-mode structural features include lesion grayscale features, lesion texture features, lesion boundary and morphological features, and contrast features; the step of extracting features from each region of the first region set to obtain the B-mode structural features of each region includes: Gray-scale statistical analysis and texture feature calculation are performed on the lesion area to obtain the gray-scale features and texture features of the lesion; Edge detection and morphological analysis are performed based on the lesion region mask and the B-mode image to extract the lesion boundary and morphological features; Gray-scale statistical analysis and texture feature calculation are performed on the adjacent reference tissue area to obtain the gray-scale features and texture features of the adjacent reference tissue area. The gray-scale features and texture features are then compared with the gray-scale features and texture features of the lesion to obtain the comparison features.

[0010] Optionally, the relative hardness distribution features include the overall relative hardness of the lesion area, the heterogeneity of hardness distribution within the lesion, and the relative hardness difference between the lesion and adjacent reference tissue areas; the step of determining a second region set of the true elastography image corresponding to the B-mode image based on the division of the first region set, and extracting hardness features from each region of the second region set to obtain the relative hardness distribution features of each region includes: Based on the division of the lesion area and the adjacent reference tissue area, the true lesion area and the true adjacent reference tissue area of ​​the true elastography image are determined respectively. The overall relative hardness of the lesion region and the heterogeneity of hardness distribution within the lesion are obtained from the actual lesion region. The hardness characteristics of the actual lesion area and the actual adjacent reference tissue area are compared and analyzed to obtain the relative hardness difference between the lesion and the adjacent reference tissue area.

[0011] Optionally, the step of performing feature mapping on the B-mode structural features and the relative hardness distribution features of the same type of lesions to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion includes: The sample data sets are clustered and grouped according to the lesion type to obtain sample data pairs for each type of lesion. Based on the correspondence of the sample data pairs, the B-mode structural features and the relative hardness distribution features are spatially registered and associated to obtain the B-mode feature-hardness distribution mapping relationship for each sample data pair. Based on the sample data, the B-mode feature-hardness distribution mapping relationship within the group is statistically mapped to obtain the prior knowledge base of B-mode feature-hardness distribution for each type of lesion.

[0012] Optionally, the implicit mechanical feature extraction of the B-mode image to be processed, based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, to obtain the tissue mechanical latent characterization of the B-mode image to be processed, includes: Obtain the structural features of the B-mode image to be processed; The mapping relationship between the structural features of the B-mode to be processed and the prior knowledge base of the B-mode features-hardness distribution is matched to obtain the potential tissue mechanical characterization of the B-mode image to be processed.

[0013] Optionally, after fusing the elastic imaging information and the image background information to obtain the virtual elastic imaging of the B-mode image to be processed, the method further includes: The credibility of the virtual elastic imaging is evaluated based on a preset credibility evaluation index. The credibility evaluation index includes at least one of the following: the quality of the B-mode image to be processed, the confidence level of lesion segmentation, the stability of B-mode structural features, the consistency of potential tissue mechanical representation, the degree of matching of prior knowledge base, the consistency of bi-branch reconstruction results, and the effect of background region suppression.

[0014] A system for constructing virtual elastic imaging, comprising: The sample acquisition module is used to acquire a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image. The B-mode image processing module is used to divide the B-mode image into regions to obtain a first region set of the B-mode image, and to extract features from each region of the first region set to obtain the B-mode structural features of each region. The real elastic imaging image processing module is used to determine the second region set of the real elastic imaging image corresponding to the B mode image based on the division of the first region set, and to extract the hardness features of each region in the second region set to obtain the relative hardness distribution features of each region. The knowledge base construction module is used to perform feature mapping on the B-mode structural features and the relative hardness distribution features of the same type of lesions, and to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. The implicit mechanical feature extraction module is used to extract implicit mechanical features from the B-mode image to be processed based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, so as to obtain the potential tissue mechanical characterization of the B-mode image to be processed. An elastography information generation module is used to generate elastography information of the B-mode image to be processed based on the potential tissue biomechanical characterization and the lesion-related region of the B-mode image to be processed. An image background information generation module is used to generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background area of ​​the B-mode image to be processed. A virtual elastic imaging module is used to fuse the elastic imaging information and the image background information to obtain a virtual elastic imaging of the B-mode image to be processed.

[0015] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain a first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. For the same type of lesion, feature mapping is performed on the B-mode structural features and the relative hardness distribution features to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related region of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed; The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

[0016] In this embodiment, by establishing a prior knowledge base of B-mode features and hardness distribution, the mapping relationship between B-mode structural features and relative hardness distribution features is fully explored. Then, the potential tissue mechanical characteristics are extracted from the B-mode image to be processed, generating elastic imaging information with accurate relative hardness distribution characteristics. Combined with the elastic imaging information generated in the lesion-related area and fused with the background information, a virtual elastic imaging of the corresponding B-mode image can be accurately constructed. This allows the virtual elastic imaging to accurately reflect the hardness differences and distribution characteristics of the tissue, combining the structural information of the B-mode image with the hardness characterization capability of real elastic imaging, thereby effectively assisting in endoscopic ultrasound research. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this embodiment, the accompanying drawings used in the description of the embodiment will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of a scenario for an embodiment of the virtual elastic imaging construction system provided by the present invention; Figure 2 A schematic diagram of another embodiment of the virtual elastic imaging construction system provided by the present invention; Figure 3 A schematic flowchart of an embodiment of the virtual elastic imaging construction method provided by the present invention; Figure 4 A schematic diagram of the structure of an embodiment of the virtual elastic imaging construction system provided by the present invention; Figure 5 A flowchart illustrating an embodiment of the invention for constructing a prior knowledge base for B-mode features and hardness distribution; Figure 6 This is a schematic diagram of the structure of an embodiment of the construction network for bi-branch virtual elastic imaging provided by the present invention; Figure 7 A schematic diagram of the structure of an embodiment of the virtual elastic imaging construction system provided by the present invention; Figure 8 This is a schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] The technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In the following description, specific embodiments of the invention will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the invention described above are not intended to be limiting, and those skilled in the art will understand that many of the steps and operations described below can also be implemented in hardware.

[0021] The terms "module" or "unit" as used herein can be considered as software objects executing on the computing system. The different components, modules, engines, and services described herein can be considered as implementation objects on the computing system. The apparatus and methods described herein are preferably implemented in software, but can also be implemented in hardware, both of which are within the scope of this invention.

[0022] This invention provides a method, system, and electronic device for constructing virtual elastic imaging.

[0023] Please see Figure 1 , Figure 1 This is a schematic diagram of a scenario illustrating an embodiment of the virtual elastic imaging construction system provided by the present invention. The virtual elastic imaging construction system may include a client 100 and a server 200, which are connected via a network. The server 200 integrates a virtual elastic imaging construction device. The server 200 may be a work platform server (i.e., a server loaded with a work platform), such as... Figure 1In this embodiment, the server 200 is mainly used to acquire a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elastography image; the B-mode image is divided into regions to obtain a first set of regions of the B-mode image, and features are extracted from each region of the first set to obtain the B-mode structural features of each region; based on the division of the first set of regions, a second set of regions of the real elastography image corresponding to the B-mode image is determined, and hardness features are extracted from each region of the second set to obtain the relative hardness distribution features of each region; the B-mode structural features and relative hardness distribution features of the same type of lesion are further analyzed. A feature mapping process is performed to establish a prior knowledge base of B-mode features and stiffness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image. Based on the potential tissue mechanical characterization and the lesion-related regions of the B-mode image to be processed, elastic imaging information of the B-mode image to be processed is generated. Based on the B-mode image to be processed and its background region, image background information of the B-mode image to be processed is generated. The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

[0024] In this embodiment, the server 200 can be a standalone server, a server network, or a server cluster. For example, the server 200 described in this embodiment includes, but is not limited to, computers, network hosts, single network servers, multiple network server sets, or cloud servers composed of multiple servers. The cloud server is composed of a large number of computers or network servers based on cloud computing. In this embodiment, communication between the server and the client can be achieved through any communication method, including but not limited to mobile communication based on the 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), and Worldwide Interoperability for Microwave Access (WiMAX), or computer network communication based on the TCP / IP Protocol Suite (TCP / IP) and User Datagram Protocol (UDP).

[0025] It is understood that the client 100 used in this embodiment can be understood as a client device. A client device includes both receiving and transmitting hardware, that is, a device with receiving and transmitting hardware capable of performing bidirectional communication on a bidirectional communication link. Such a client device may include: cellular or other communication devices, having a single-line display, a multi-line display, or a cellular or other communication device without a multi-line display. Specifically, the client 100 may be a desktop terminal or a mobile terminal, specifically a mobile phone, tablet computer, laptop computer, etc.

[0026] Those skilled in the art will understand that Figure 1 The application environment shown is merely one application scenario of the solution in this application and does not constitute a limitation on the application scenario of the solution in this application. Other application environments may include those that are more specific to this application. Figure 1 The number of more or fewer servers shown, or the server network connectivity relationships, for example... Figure 1 Only one server and two clients are shown in the diagram. It is understood that the virtual elastic imaging construction system may also include one or more other servers, and / or one or more clients connected to the server network, which is not limited here.

[0027] In some embodiments of the present invention, the working platform may be an enterprise office platform, such as WeChat for Business. Taking server 200 as an example, it may further include an enterprise office platform contact server, an enterprise office platform configuration management server, and a web management server. Enterprise users or developers can access the web management server using a web browser terminal to configure the field configuration information on the enterprise office platform configuration management server, and set and store the enterprise user information of enterprise employees of the enterprise office platform on the enterprise office platform contact server.

[0028] In addition, such as Figure 2 As shown, Figure 2 This is a schematic diagram of another embodiment of the virtual elastic imaging construction system provided by the present invention. The virtual elastic imaging construction system may further include a storage terminal 300 for storing data, such as a storage object database. The object database stores object data, which may include application templates (such as approval templates, attendance templates, and other application templates), file data (such as Word files, Excel files, or PPT files, and other files in various formats), image data (such as images in various formats such as jpg, png, and bmp), and other types of data. Correspondingly, the object database may also be divided into multiple types of data, such as an application database, a file database, or an image database.

[0029] It should be noted that, Figure 1-2The schematic diagram of the virtual elastic imaging construction system shown is merely an example. The virtual elastic imaging construction system and scenario described in this embodiment are for the purpose of more clearly illustrating the technical solution of this embodiment and do not constitute a limitation on the technical solution provided by this invention. As those skilled in the art will know, with the evolution of the virtual elastic imaging construction system and the emergence of new business scenarios, the technical solution provided by this invention is also applicable to similar technical problems.

[0030] The following detailed description is based on specific embodiments.

[0031] In this embodiment, the description will be from the perspective of a virtual elastic imaging construction device, which can be integrated into the server 200.

[0032] This invention provides a method for constructing virtual elastic imaging; please refer to [link / reference]. Figure 3 , Figure 3 A flowchart illustrating an embodiment of the virtual elastic imaging construction method provided by the present invention includes: S301: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; S302: Divide the B-mode image into regions to obtain the first region set of the B-mode image, and extract features from each region of the first region set to obtain the B-mode structural features of each region. S303: Based on the division of the first region set, determine the second region set of the real elastic imaging image corresponding to the B mode image, and extract the hardness features of each region in the second region set to obtain the relative hardness distribution features of each region. S304: Perform feature mapping on the B-mode structural features and relative hardness distribution features of the same type of lesions, and establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. S305: Based on the mapping relationship in the prior knowledge base of B-mode features and hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. S306: Generate elastography information of the B-mode image to be processed based on the potential biomechanical characterization and the lesion-related region of the B-mode image to be processed; S307: Generate the image background information of the B-mode image to be processed based on the B-mode image to be processed and the background area of ​​the B-mode image to be processed; S308: The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

[0033] In one specific embodiment, the sample data pair set refers to a screened and reliable training dataset, which consists of high-quality B-mode images paired with corresponding real elastic imaging images. These data pairs have a uniform format and spatial alignment relationship, and are the basis for constructing a prior knowledge base of B-mode features-stiffness distribution and training a model for virtual elastic imaging.

[0034] B-mode images are the most commonly used basic image format in endoscopic ultrasound examinations. They can display the location, size, shape, boundary, echo intensity, internal echo distribution, and relationship with surrounding tissues of lesions. They reflect the echo characteristics of tissues through different grayscale values ​​and are core imaging data in clinical endoscopic ultrasound examinations.

[0035] True elastography images are acquired directly using elastography mode during endoscopic ultrasound examinations. They reflect mechanical information such as tissue stiffness, tissue strain, and compressive compliance. They use color coding to represent the relative stiffness of the tissue, typically with blue indicating harder tissue and red indicating softer tissue.

[0036] The first region set is the set of regions obtained after dividing the B-mode image into regions. It generally includes the lesion region, the adjacent reference tissue region, and the background tissue region. It can also include only the lesion region or the background tissue region according to actual needs to achieve targeted feature extraction. There is no restriction here.

[0037] B-mode structural features are image proxy features extracted from B-mode images that are related to the mechanical state of the tissue. They reflect information such as grayscale intensity, echo uniformity, texture heterogeneity, boundary clarity, and morphological regularity of lesions, and are used to infer the mechanical properties of the tissue.

[0038] The second region set is the set of regions corresponding to the true elastic imaging image determined based on the first region set division of the B-mode image. It corresponds completely to the region division result of the first region set.

[0039] Relative hardness distribution features are characteristics extracted from real elastography images that reflect the distribution of tissue hardness. They are used to characterize the overall elasticity of the lesion (whether it is relatively hard or soft), the consistency or difference of the hardness distribution within the lesion, and the hardness relationship of the lesion relative to the adjacent reference tissue area.

[0040] The same type of lesion refers to lesions with the same pathological type or histological origin, or the same lesion from the same patient. When constructing the B-mode feature-hardness distribution prior knowledge base, it is necessary to establish the correspondence between the B-mode structural features and the relative hardness distribution features for the same type of lesion to ensure the accuracy and reliability of feature mapping.

[0041] The B-mode feature-hardness distribution prior knowledge base is a set of correspondences between B-mode structural features and the hardness distribution in real elastography, established based on paired training data. It is used to characterize the overall correspondence between different B-mode structural features and the relative hardness distribution features in real elastography, thereby providing data support for subsequent inference of tissue mechanical properties based on B-mode images.

[0042] The B-mode image to be processed is the ultrasound endoscopic B-mode image that needs to be constructed for virtual elastography. It can be a single static image or a keyframe extracted from the ultrasound endoscopic video, and there is no restriction here.

[0043] Tissue biomechanical latent characterization is an individualized implicit representation of the mechanical characteristics generated by the current lesion. It is used to characterize the relative mechanical relationships that the current lesion may have. It is not a direct measurement of the elastic modulus of the real tissue, but is inferred based on the prior information of the structural features and stiffness distribution of the current lesion in B-mode.

[0044] The lesion-related region of the B-mode image to be processed refers to the image region segmented from the B-mode image to be processed that is directly related to the lesion assessment. Generally, it can be the lesion region directly identified from the B-mode image to be processed, or it can be a composite region that includes the adjacent reference tissue region, which is expanded based on the identified lesion region. There are no restrictions here, as long as the needs of subsequent feature extraction and mechanical inference are met.

[0045] Elastography information is virtual elastic information generated based on the potential characterization of tissue mechanics and the lesion-related region of the B-mode image to be processed. It includes information such as the stiffness distribution, strain differences and mechanical heterogeneity within the lesion region of the B-mode image to be processed, reflecting the virtual elastic performance of the lesion region and adjacent reference tissue areas, and is the core component of virtual elastography images.

[0046] The background region of the B-mode image to be processed is the image region other than the lesion-related region, which is used to maintain the image background structure and suppress the generation of elastic information that is not clinically meaningful.

[0047] Image background information is background tissue environment information generated based on the B-mode image to be processed and the background region of the B-mode image to be processed. It is used to maintain the consistency of the tissue environment, spatial structure and image background around the lesion.

[0048] Virtual elastography is a generated elastography image corresponding to the B-mode image to be processed. It integrates elastography information and image background information, and can provide auxiliary imaging information for lesion assessment under endoscopic ultrasound in the case of elastography without acquiring or saving the B-mode image to be processed.

[0049] In this embodiment, by establishing a prior knowledge base of B-mode features and hardness distribution, the mapping relationship between B-mode structural features and relative hardness distribution features is fully explored. Then, the potential tissue mechanical characteristics are extracted from the B-mode image to be processed, generating elastic imaging information with accurate relative hardness distribution characteristics. Combined with the elastic imaging information generated in the lesion-related area and fused with the background information, a virtual elastic imaging of the corresponding B-mode image can be accurately constructed. This allows the virtual elastic imaging to accurately reflect the hardness differences and distribution characteristics of the tissue, combining the structural information of the B-mode image with the hardness characterization capability of real elastic imaging, thereby effectively assisting in endoscopic ultrasound research.

[0050] In one specific embodiment, in S301, obtaining a set of sample data pairs includes: screening initial sets of sample data pairs under the same case, the same examination process, the same lesion, or the same image scene; removing abnormal data pairs, incomplete data pairs, and artifact data pairs from the initial set of sample data pairs to obtain a first set of sample data pairs; performing standardized preprocessing and spatial registration on the first set of sample data pairs to obtain a set of sample data pairs; wherein, the standardized preprocessing includes at least one of image cropping, size normalization, grayscale normalization, brightness and contrast correction, noise suppression, removal of irrelevant text or borders, image orientation unification, color space conversion, and image format conversion; spatial registration includes at least one of deep learning, affine transformation, or non-rigid deformation.

[0051] The initial sample data set is obtained by selecting paired data from the same case, the same examination process, the same lesion, or the same image scene, based on the paired data of B-mode images and real elastography images that can be directly obtained by collecting existing databases. These data pairs may contain various quality problems or format differences and are the basis for building a high-quality training dataset.

[0052] Abnormal data pairs refer to data pairs in the initial sample data pair set that have serious quality problems, including but not limited to image pairs with severe blur, overexposure, underexposure, obvious artifacts, incomplete lesion display, unidentifiable lesions, unstable image freezing, or incomplete acquisition information. These data pairs will affect the accuracy and stability of model training and need to be removed during the quality control stage.

[0053] Incomplete data pairs refer to data pairs in the initial sample data pair set that lack necessary information, such as image pairs that lack lesion display, missing key anatomical structures, or incomplete acquisition information. These data pairs cannot provide enough effective information for model training and need to be identified and removed during the quality control stage.

[0054] Artifact data pairs refer to data pairs in the initial sample data pair set that contain obvious artifacts, such as motion artifacts, breathing artifacts, instrument artifacts, etc. These artifacts can interfere with the extraction and analysis of image features and affect the model's learning of real tissue features, and need to be removed during the quality control stage.

[0055] The first sample data pair set refers to the high-quality data pair set obtained after quality control, which removes outlier, incomplete, and artifact data pairs. These data pairs have good image quality and integrity, providing a reliable foundation for subsequent standardized preprocessing and model training.

[0056] Standardization preprocessing refers to a series of uniform processing steps performed on the images in the first sample data set to eliminate format and quality differences between different images. This includes image cropping, size normalization, grayscale normalization, brightness and contrast correction, noise suppression, removal of irrelevant text or borders, image orientation unification, color space conversion, and image format conversion. Standardization preprocessing can improve the stability and accuracy of model training.

[0057] Spatial registration refers to the alignment process between a standardized B-mode image and a true elastic imaging image to ensure their consistency in spatial location.

[0058] Image cropping is an important step in standardized preprocessing. It refers to removing irrelevant borders, text, or other non-anatomical structures from an image, retaining only the effective areas containing lesions and key anatomical structures.

[0059] Size normalization refers to adjusting images of different sizes to the same size to eliminate the impact of image size differences on model training. Size normalization is usually implemented using interpolation algorithms, which can ensure the consistency of model input and improve the stability of model training.

[0060] Gray-scale normalization refers to adjusting images with different gray-scale ranges to a uniform gray-scale range to eliminate the impact of gray-scale differences on model training. Gray-scale normalization is usually achieved through histogram equalization or linear transformation, which can improve the comparability of image features and the accuracy of model training.

[0061] Brightness and contrast correction refers to adjusting the brightness and contrast of an image to improve its visual effect and the accuracy of feature extraction. Brightness correction can eliminate overexposure or underexposure problems, while contrast correction can enhance the differences between different tissues in an image.

[0062] Noise suppression refers to removing noise interference from an image to improve image quality and the accuracy of feature extraction. Noise suppression is usually achieved using filtering algorithms, which can reduce the impact of noise on model training and improve the model's ability to learn real tissue features.

[0063] Irrelevant text or border removal refers to removing text labels, measurement rulers, or other non-anatomical structure borders from an image to reduce interference from irrelevant information. This irrelevant information can affect the extraction and analysis of image features; removing it can improve the efficiency and accuracy of model training.

[0064] Image orientation unification refers to adjusting images with different orientations to a unified orientation to eliminate the impact of image orientation differences on model training. Image orientation unification is usually based on the position of anatomical structures, which can ensure the consistency of model input and improve the stability of model training.

[0065] Color space conversion refers to transforming an image from one color space to another to meet the needs of model training. For example, converting a color image to a grayscale image, or a grayscale image to a pseudo-color image. Color space conversion can improve the expressive power of image features and enhance the model's ability to learn different organizational features.

[0066] Image format conversion refers to converting images of different formats into a unified format to eliminate the impact of image format differences on model training. Image format conversion typically converts images to common formats such as JPEG, PNG, or DICOM, ensuring consistency of model input and improving the stability of model training.

[0067] Deep learning refers to machine learning methods based on deep neural networks, which can automatically learn complex feature representations from data. Deep learning can be used in multiple stages such as image registration, lesion segmentation, feature extraction, inference of potential tissue mechanical characteristics, and construction of virtual elastography. It is one of the core technologies for realizing the construction of virtual elastography.

[0068] Affine transformation is a linear transformation that can perform operations such as translation, rotation, scaling, and shearing on images. It can be used for image spatial registration to align B-mode images with real elastic imaging images. Affine transformations are computationally simple and fast, and are suitable for rigid or near-rigid image registration.

[0069] Non-rigid deformation refers to a non-linear transformation that enables arbitrary shape changes in images. It can be used for image spatial registration to achieve precise alignment between B-mode images and true elastic imaging images, especially in the presence of tissue deformation. Non-rigid deformation can handle more complex image registration problems and improve registration accuracy.

[0070] In this embodiment, by screening initial sample data pairs from the same source scenario, such as the same case and the same examination process, and removing abnormal, incomplete, and artifact data pairs, and combining standardized operations covering multiple preprocessing methods with spatial registration including deep learning and affine transformation, the homology, validity, and consistency of sample data pairs are effectively guaranteed, the sample quality is greatly improved, and a precise and reliable data foundation is provided for the subsequent construction of virtual elastic imaging, thereby ensuring the accuracy and reliability of the reconstruction results.

[0071] Furthermore, in practical scenarios, the process of acquiring and preprocessing sample data sets specifically includes: First, B-mode images and their corresponding true elastography images obtained during endoscopic ultrasound examinations were collected to construct a raw image dataset. Endoscopic ultrasound images can originate from lesions of the biliary and pancreatic system, submucosal tumors of the gastrointestinal tract, perigastrointestinal lymph nodes, or other lesions observable under endoscopic ultrasound.

[0072] Then, B-mode images and real elastography images from the same case, the same examination procedure, the same lesion, or the same image scene are selected to establish an initial pairing relationship. Specifically, the real elastography images are derived from the elastography pattern of the same endoscopic ultrasound examination. In particular, for elastography images displayed on both sides, the B-mode display area and the elastography display area in the image can be separated to obtain the initial B-mode image and the initial elastography image. The initial B-mode image is used for subsequent structural feature extraction and model input, while the initial elastography image is used as the supervised target for subsequent stiffness distribution prior construction and virtual elastography construction training.

[0073] Secondly, quality control is performed on the acquired images. Quality control methods include, but are not limited to, assessments of image sharpness, brightness, contrast, artifact severity, lesion display completeness, lesion identifiability, image freeze-thaw stability, and the completeness of image acquisition information. Images that are severely blurred, overexposed, underexposed, have obvious artifacts, incomplete lesion display, unidentifiable lesions, unstable image freeze-thaw, or incomplete acquisition information are discarded.

[0074] Next, the images that have passed quality control are subjected to standardized preprocessing. The standardized preprocessing methods include, but are not limited to, image cropping, size normalization, grayscale normalization, brightness and contrast correction, noise suppression, removal of irrelevant text or borders, image orientation unification, color space conversion, and image format conversion.

[0075] Finally, spatial registration is performed on the standardized B-mode images and standardized elastography images to obtain a set of sample data pairs. During the registration process, registration methods based on lesion regions, key points, image similarity, affine transformation, non-rigid transformation, or deep learning can be used to form B-mode image-real elastography image pairing data for subsequent training.

[0076] In one specific embodiment, in S302, the first region set includes a lesion region, an adjacent reference tissue region, and a background tissue region; dividing the B-mode image into regions to obtain the first region set of the B-mode image includes: performing target recognition on the B-mode image to obtain a lesion region mask, and determining the lesion region based on the lesion region mask; determining the region in the B-mode image that is at a preset distance from the lesion region as the adjacent reference tissue region; and determining the remaining region in the B-mode image other than the lesion region and the adjacent reference tissue region as the background tissue region.

[0077] It should be noted that the lesion area refers to the region where the target lesion is located in the B-mode image of endoscopic ultrasound, which is the core area of ​​focus in the construction of virtual elastography.

[0078] A lesion region mask is a binary image obtained after target recognition of a B-mode image through a lesion segmentation network, used to mark the location and extent of the lesion region.

[0079] In real-world scenarios, to ensure the segmentation accuracy of the first region set, for B-mode images, physicians with experience in endoscopic ultrasound diagnosis annotate the target lesion regions, forming a lesion region annotation dataset. Annotation methods include, but are not limited to, lesion bounding box annotation, lesion contour annotation, and lesion mask annotation.

[0080] Based on the obtained lesion region annotation dataset, a lesion segmentation network is constructed, and the B-mode image is input into the lesion segmentation network for target recognition, thereby obtaining a lesion region mask. Based on the lesion region mask, the lesion region, adjacent reference tissue region, and background tissue region are determined. Specifically, the lesion region is the area where the target lesion is located, as identified by the lesion segmentation network, and is used for lesion B-mode structural feature extraction, tissue mechanical potential characterization inference, and virtual elasticity information reconstruction. The adjacent reference tissue region is the tissue region surrounding the lesion region that can serve as a reference for relative hardness comparison. The adjacent reference tissue region can be obtained by extending the lesion region mask outward by a preset range, or by physician annotation, image segmentation model, attention model, or other region recognition methods. The background tissue region is other image regions besides the lesion region and adjacent reference tissue region, used to maintain the image background structure and suppress the generation of clinically meaningless elasticity information.

[0081] The lesion segmentation network can be U-Net, U-Net++, DeepLab, Mask R-CNN, Transformer segmentation model, visual attention model, interactive segmentation model, or other artificial intelligence models that can achieve lesion localization and segmentation.

[0082] Furthermore, to improve the reliability of the lesion segmentation network, region constraint information can be constructed for subsequent B-mode structural feature extraction, inference of potential tissue biomechanical representations, construction of virtual elastography, and background region suppression. This region constraint information includes one or more of the following: lesion region mask, adjacent reference tissue region mask, background tissue region mask, and corresponding region weights.

[0083] The adjacent reference tissue area refers to the tissue region located around the lesion area that can serve as a reference for relative hardness comparison. The selection of the adjacent reference tissue area needs to take into account both anatomical rationality and image feature stability. It usually covers a certain range extending beyond the edge of the lesion, aiming to provide a control benchmark with similar acoustic characteristics to the lesion but different pathological states. The specific selection range can be flexibly set according to the size of the lesion and the surrounding anatomical structures, such as 1 to 2 times the diameter of the lesion, or adaptively adjusted according to specific anatomical boundaries (such as the fascia layer, organ capsule), to ensure that the reference area is not directly affected by the lesion infiltration and can truly reflect the mechanical properties of the surrounding normal tissue.

[0084] It should be noted that, in one specific embodiment, to facilitate calculations by the electronic device, a preset distance is set to achieve automated extraction of the adjacent reference tissue area. The preset distance refers to the range of distances extended outward from the lesion area when determining the adjacent reference tissue area. This preset distance can be adjusted according to the lesion type, lesion size, and clinical needs, and is used to determine the range of adjacent tissue areas that can serve as a reference for relative hardness comparison. Generally, a preset distance of 3-5 mm is preferred. A reasonable setting of the preset distance ensures that the selection of the adjacent reference tissue area is clinically significant, providing a reliable reference basis for inferring the relative hardness of the lesion.

[0085] Background tissue area refers to the image region other than the lesion area and adjacent reference tissue area, which is used to maintain the image background structure and suppress the generation of elastic information that is not clinically meaningful.

[0086] In this embodiment, the lesion area is accurately determined by target recognition of the B-mode image. Then, the adjacent reference tissue area is delineated by a preset distance, and the remaining area is set as the background tissue area. This achieves accurate hierarchical division of the B-mode image, providing a clear and specific regional basis for the subsequent construction of virtual elastography. This can improve the ability of elastography to distinguish between lesions and different types of tissues, thereby improving the accuracy and reliability of the construction of virtual elastography and helping clinicians to accurately judge tissue characteristics.

[0087] In one specific embodiment, the B-mode structural features include lesion grayscale features, lesion texture features, lesion boundary and morphological features, and contrast features. Feature extraction is performed on each region of the first region set to obtain the B-mode structural features of each region, including: performing grayscale statistical analysis and texture feature calculation on the lesion region to obtain lesion grayscale features and lesion texture features; performing edge detection and morphological analysis based on the lesion region mask and the B-mode image to extract lesion boundary and morphological features; performing grayscale statistical analysis and texture feature calculation on the adjacent reference tissue region to obtain the grayscale features and texture features of the adjacent reference tissue region, and comparing the grayscale features and texture features with the lesion grayscale features and lesion texture features respectively to obtain contrast features.

[0088] Among them, lesion grayscale features refer to the grayscale intensity-related features extracted from the lesion area of ​​endoscopic ultrasound B-mode images, including one or more of the following: mean grayscale, median grayscale, standard deviation of grayscale, and coefficient of variation of grayscale. Lesion grayscale features are used to characterize the echogenicity and tissue heterogeneity within lesions, reflecting the possible differences in hardness and mechanical heterogeneity within the lesion, and are one of the important imaging proxy features for inferring the mechanical state of tissues.

[0089] Lesion texture features refer to texture-related features extracted from lesion regions in endoscopic ultrasound B-mode images, including one or more of echo homogeneity, texture roughness, and texture entropy. These features characterize echo heterogeneity, local texture complexity, and spatial heterogeneity within lesions, reflecting potential hardness differences and mechanical heterogeneity within the lesion, and are one of the important imaging proxy features for inferring tissue mechanical state.

[0090] Lesion boundary and morphological features refer to the features related to the lesion boundary and morphology extracted based on the lesion region mask and the corresponding B-mode image, including one or more of boundary sharpness, boundary continuity, and morphological regularity. These features are used to characterize the edge appearance and overall morphology of the lesion, reflect the possible hardness transition pattern between the lesion and surrounding tissues, and are one of the important imaging proxy features for inferring the tissue mechanical state.

[0091] Contrast features are features obtained by comparing the grayscale and texture features of a neighboring reference tissue area with the corresponding features of the lesion area. These features include one or more of grayscale differences, grayscale ratios, and texture differences. These features are used to provide a reference for inferring the relative hardness difference between the lesion and the neighboring reference tissue area, reflecting the hardness relationship of the lesion relative to the adjacent tissue, and are one of the important imaging proxy features for inferring the mechanical state of tissues.

[0092] Grayscale statistical analysis refers to the statistical analysis of grayscale intensity in lesion areas of endoscopic ultrasound B-mode images to extract grayscale features of the lesions. By calculating statistical indicators such as mean grayscale, median grayscale, standard deviation of grayscale, and coefficient of variation of grayscale, the echogenicity and tissue heterogeneity within the lesions can be quantified, providing important evidence for inferring the tissue's mechanical state.

[0093] Texture feature calculation refers to the process of calculating and analyzing the texture representation of lesion areas in B-mode ultrasound images to extract lesion texture features. By calculating texture indices such as echo uniformity, texture roughness, and texture entropy, the echo heterogeneity, local texture complexity, and spatial heterogeneity within the lesion can be quantified, providing important evidence for inferring the tissue mechanical state.

[0094] Edge detection and morphological analysis refers to the process of extracting the boundary and morphological features of a lesion by performing edge detection and morphological analysis based on a lesion region mask and the corresponding B-mode image. By analyzing indicators such as boundary clarity, boundary continuity, and morphological regularity, the edge performance and overall morphology of the lesion can be quantified, reflecting the possible hardness transition patterns between the lesion and surrounding tissues, and providing important evidence for inferring the tissue's mechanical state.

[0095] Grayscale statistical analysis and texture feature calculation refer to the statistical analysis of grayscale intensity and the calculation of texture representation of adjacent reference tissue areas in endoscopic ultrasound B-mode images, in order to extract grayscale and texture features of the adjacent reference tissue areas. By calculating indicators such as mean grayscale, median grayscale, standard deviation grayscale, coefficient of variation grayscale, echo uniformity, texture roughness, and texture entropy, the echo representation and tissue heterogeneity of adjacent reference tissue areas can be quantified, providing a reference basis for inferring the relative hardness difference between lesions and adjacent reference tissue areas.

[0096] In this embodiment, by performing grayscale statistical analysis, texture feature calculation, edge detection, and morphological analysis on the lesion area, the grayscale features, texture features, and boundary and morphological features of the lesion are accurately extracted. The lesion features are then compared with the corresponding features of the adjacent reference tissue area to obtain comparative features. This allows for the comprehensive and multi-dimensional acquisition of B-mode structural features, providing rich and accurate structural basis for the construction of virtual elastography, effectively improving the accuracy and reliability of elastography reconstruction, and helping to achieve accurate analysis and evaluation of lesions.

[0097] In practical scenarios, based on the acquired lesion area and adjacent reference tissue area, B-mode structural features are extracted from the B-mode image; simultaneously, background tissue environment information is acquired for the background tissue area. The B-mode structural features include lesion grayscale features, lesion texture features, lesion boundary and morphological features, and contrast features between the lesion and adjacent reference tissue areas; the background tissue environment information is used for subsequent background tissue structure preservation and suppression of ineffective elasticity information generation, and is not used as a direct basis for inferring lesion hardness distribution.

[0098] Specifically, for the lesion area, grayscale features and texture features of the lesion are extracted. The grayscale features of the lesion include one or more of the following: mean grayscale, median grayscale, standard deviation of grayscale, and coefficient of variation of grayscale; the texture features of the lesion include one or more of the following: echo uniformity, texture roughness, and texture entropy, which are used to characterize the echo appearance and tissue heterogeneity within the lesion.

[0099] Lesion boundary and morphological features are extracted based on a lesion region mask and the corresponding B-mode image. These features include one or more of the following: boundary sharpness, boundary continuity, and morphological regularity, used to characterize the edge appearance and overall morphology of the lesion.

[0100] For adjacent reference tissue areas, grayscale and texture features are extracted and compared with corresponding features of the lesion area to obtain contrast features between the lesion and adjacent reference tissue areas. Contrast features include one or more of grayscale difference, grayscale ratio, and texture difference, which are used to provide a reference for inferring the relative hardness difference between the lesion and adjacent reference tissue areas.

[0101] For the background tissue region, background tissue environment information is acquired for subsequent background tissue structure preservation and suppression of invalid elasticity information generation in the background region. Background tissue environment information is not used as a direct basis for inferring lesion hardness.

[0102] In one specific embodiment, in S303, the relative hardness distribution features include the overall relative hardness of the lesion area, the heterogeneity of hardness distribution within the lesion, and the relative hardness difference between the lesion and the adjacent reference tissue area. Based on the division of the first region set, a second region set of the true elastography image corresponding to the B-mode image is determined, and hardness features are extracted from each region of the second region set to obtain the relative hardness distribution features of each region, including: determining the true lesion area and the true adjacent reference tissue area of ​​the true elastography image based on the division of the lesion area and the adjacent reference tissue area respectively; obtaining the overall relative hardness of the lesion area and the heterogeneity of hardness distribution within the lesion; and comparing and analyzing the hardness features of the true lesion area and the true adjacent reference tissue area to obtain the relative hardness difference between the lesion and the adjacent reference tissue area.

[0103] The overall relative hardness of the lesion area refers to the features extracted from real elastography images that reflect the overall hardness of the lesion, used to characterize the overall elastic performance of the lesion as being either too hard or too soft. Specifically, it can be obtained from color coding, relative hardness mapping, regional hardness characterization values, elasticity scores, or other information that can characterize the distribution of relative hardness in real elastography images. It does not represent a direct measurement of the elastic modulus of real tissue, but is used to guide the inference of the potential mechanical characterization of the current lesion tissue and to constrain the construction results of virtual elastography to conform to the tissue hardness distribution law.

[0104] Heterogeneity of hardness distribution within a lesion refers to the features extracted from real elastography images that reflect the differences in hardness distribution within the lesion. It is used to characterize the consistency or difference in hardness distribution within a lesion. It can be obtained based on color coding, relative hardness mapping, regional hardness characterization values, elasticity scores, or other information that can characterize relative hardness distribution in real elastography images. It does not represent a direct measurement of the elastic modulus of real tissue. It is used to guide the inference of the potential mechanical characterization of the current lesion tissue and to constrain the construction results of virtual elastography to conform to the tissue hardness distribution law.

[0105] The relative stiffness difference between the lesion and the adjacent reference tissue area refers to the feature extracted from the real elastography image that reflects the stiffness relationship between the lesion and the adjacent reference tissue area. It is used to characterize whether the lesion is harder or softer than the adjacent reference tissue area. It can be obtained from the color coding, relative stiffness mapping, regional stiffness characterization value, elasticity score or other information that can characterize the relative stiffness distribution in the real elastography image. It does not represent a direct measurement of the elastic modulus of the real tissue. It is used to guide the inference of the potential mechanical characterization of the current lesion tissue and to constrain the construction results of the virtual elastography to conform to the tissue stiffness distribution law.

[0106] The true lesion area refers to the lesion area division based on the B-mode image. The lesion area corresponding to the true elastography image is determined and corresponds to the lesion area in the B-mode image. It is used to extract the relative stiffness distribution features in the true elastography image and establish the correspondence between the B-mode structural features and the true elastography stiffness distribution.

[0107] The true neighboring reference tissue region refers to the neighboring reference tissue region division based on the B-mode image. It is determined to be the neighboring reference tissue region corresponding to the true elastic imaging image, which corresponds to the neighboring reference tissue region in the B-mode image. It is used to extract the relative stiffness distribution features in the true elastic imaging image and establish the correspondence between the B-mode structural features and the stiffness distribution of the true elastic imaging.

[0108] In this embodiment, by defining three specific indicators of relative hardness distribution characteristics, and determining the relevant regions of the real elastic imaging image based on the region division of the B-mode image, the overall relative hardness of the lesion, the heterogeneity of internal hardness distribution, and the relative hardness difference between the lesion and the adjacent reference tissue area are accurately extracted and analyzed. This allows for comprehensive and accurate acquisition of the hardness characteristic information of the lesion, providing a more reliable basis for hardness analysis for subsequent medical diagnosis and other work.

[0109] In one specific embodiment, in S304, feature mapping is performed on the B-mode structural features and relative hardness distribution features of the same type of lesions to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. This includes: clustering the sample data pairs according to the lesion type to obtain sample data pairs for each type of lesion; based on the correspondence of the sample data pairs, spatial registration and association mapping are performed on the B-mode structural features and relative hardness distribution features to obtain the B-mode feature-hardness distribution mapping relationship for each sample data pair; and statistical mapping is performed on the B-mode feature-hardness distribution mapping relationship within the sample data pairs to obtain a prior knowledge base of B-mode features-hardness distribution for each type of lesion.

[0110] In this invention, lesion type refers to the classification of lesions based on their pathological characteristics, histological origin, or clinical manifestations, such as biliary and pancreatic system lesions, submucosal tumors of the gastrointestinal tract, and peri-gastrointestinal lymph nodes. The lesion type is used to cluster and group sample data sets to enhance the stability of the prior constraints on hardness distribution under different diseases and imaging manifestations.

[0111] Clustering refers to the process of grouping sample data according to lesion type. That is, sample data pairs with the same or similar lesion types are grouped together, thereby establishing targeted prior knowledge for different types of lesions and improving the accuracy and stability of the prior knowledge base. Through clustering, it is ensured that sample data pairs within the same group have similar B-mode structural characteristics and hardness distribution patterns, thus providing a reliable foundation for subsequent statistical mapping.

[0112] A sample data pair group refers to a collection of sample data pairs of the same lesion type obtained through clustering. It consists of multiple B-mode images with the same or similar lesion types paired with their corresponding real elastography images, and is used to establish targeted prior knowledge for different types of lesions. By statistically mapping the B-mode feature-hardness distribution mapping relationship within the sample data pair group, a prior knowledge base of B-mode feature-hardness distribution for that lesion type can be obtained.

[0113] Spatial registration and correlation mapping refers to the process of spatially aligning and correlating B-mode structural features with relative hardness distribution features based on the correspondence between sample data pairs. Through spatial registration and correlation mapping, the consistency of the spatial location of B-mode structural features and relative hardness distribution features can be ensured, providing an accurate basis for subsequent statistical mapping.

[0114] The B-pattern feature-hardness distribution mapping relationship refers to the correspondence between B-pattern structural features and relative hardness distribution features in a single sample data pair obtained through spatial registration and correlation mapping. Furthermore, by statistically mapping the B-pattern feature-hardness distribution mapping relationship of multiple sample data pairs, a priori knowledge base of B-pattern feature-hardness distribution for this lesion type can be obtained.

[0115] Statistical mapping refers to the process of statistically analyzing and summarizing the mapping relationship between B-mode features and stiffness distribution within groups based on sample data, in order to obtain a prior knowledge base of B-mode features and stiffness distribution for this lesion type. Through statistical mapping, a prior knowledge base of B-mode features and stiffness distribution for this lesion type can be obtained to guide the inference of the potential biomechanical characteristics of the current lesion and to constrain the construction results of virtual elastography to conform to the laws of tissue stiffness distribution.

[0116] In real-world scenarios, based on a set of sample data, we extract the B-mode structural features and the relative stiffness distribution features in real elastography images corresponding to the same type of lesion, and establish the correspondence between the B-mode structural features and the stiffness distribution in real elastography images.

[0117] The structural features of the B-mode are the extracted structural features related to the mechanical state of the lesion, including grayscale features of the lesion, texture features of the lesion, boundary and morphological features of the lesion, and contrast features between the lesion and the adjacent reference tissue area.

[0118] The relative hardness distribution characteristics include the overall relative hardness of the lesion area, the heterogeneity of hardness distribution within the lesion, and the relative hardness difference between the lesion and adjacent reference tissue areas. The overall relative hardness of the lesion area characterizes whether the lesion is generally harder or softer in elasticity; the heterogeneity of hardness distribution within the lesion characterizes the consistency or difference in hardness distribution within the lesion; and the relative hardness difference between the lesion and adjacent reference tissue areas characterizes whether the lesion is harder or softer relative to the adjacent reference tissue areas. These relative hardness distribution characteristics can be obtained from color coding, relative hardness mapping, regional hardness characterization values, elasticity scores, or other information that can characterize relative hardness distribution in real elastography images, and do not represent a direct measurement of the elastic modulus of real tissue.

[0119] Based on the correspondence between B-mode structural features and relative hardness distribution features in real elastography images, a B-mode feature-hardness distribution prior knowledge base is established. This prior knowledge base characterizes the correspondence between different B-mode structural features and relative hardness distribution features in real elastography. The B-mode feature-hardness distribution prior knowledge base outputs prior hardness distribution information corresponding to the current lesion based on its B-mode structural features. The prior hardness distribution information includes one or more of the following: the overall relative hardness of the lesion region, the heterogeneity of hardness distribution within the lesion, and the relative hardness relationship between the lesion and adjacent reference tissue areas. This information can be represented in the form of continuous numerical values, discrete levels, probability distributions, feature vectors, or spatial feature maps. Specifically, the prior hardness distribution information guides the inference of the potential mechanical characteristics of the lesion tissue in the current B-mode image to be processed, and also applies prior hardness distribution constraints to the construction results of the virtual elastography.

[0120] The prior knowledge base can be constructed using statistical rules, expert experience rules, machine learning models, deep representation learning models, visual language models, or combinations thereof. Furthermore, the prior knowledge base can be supplemented with information related to lesion type, lesion location, pathological results, and the degree of tissue fibrosis, or expert experience rules, to enhance the stability of the prior constraints on stiffness distribution under different diseases and imaging manifestations.

[0121] In this embodiment, by clustering sample data pairs according to lesion type, spatial registration, correlation mapping, and statistical mapping of B-mode structural features and relative hardness distribution features are completed. A prior knowledge base of B-mode features-hardness distribution for each lesion is constructed, which can provide lesion type-specific and accurate prior support for the construction of virtual elastography. This effectively improves the matching degree between the hardness distribution of the reconstructed lesion and the actual structural features, enhances the accuracy and reliability of the reconstruction results, and helps to achieve more accurate lesion assessment and diagnosis.

[0122] In one specific embodiment, in S305, based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. This includes: obtaining the structural features of the B-mode image to be processed; matching the structural features of the B-mode image to be processed with the mapping relationship in the prior knowledge base of B-mode features-hardness distribution to obtain the potential tissue mechanical characterization of the B-mode image to be processed.

[0123] Among them, the structural features of the B-mode to be processed refer to the image proxy features related to the tissue mechanical state extracted from the B-mode ultrasound image to be processed, including but not limited to the grayscale features of the lesion, the texture features of the lesion, the boundary and morphological features of the lesion, and the contrast features between the lesion and the adjacent reference tissue area.

[0124] The mapping relationship in the B-mode feature-hardness distribution prior knowledge base refers to the correspondence between B-mode structural features and the hardness distribution in real elastography, established based on paired B-mode images and real elastography images. It is obtained by feature mapping of B-mode structural features and relative hardness distribution features for the same type of lesion, and is used to characterize the overall correspondence between different B-mode structural features and relative hardness distribution features in real elastography.

[0125] In practical scenarios, in order to obtain the potential characterization of tissue mechanics, a potential characterization inference model is first constructed based on the B-mode structural features and the B-mode feature-hardness distribution prior knowledge base. Then, the B-mode structural features of the current lesion are input into the potential characterization inference model of tissue mechanics, and the potential characterization of tissue mechanics for the current lesion is obtained according to the hardness distribution prior information in the B-mode feature-hardness distribution prior knowledge base corresponding to the current lesion B-mode structural features of the B-mode image to be processed.

[0126] Among them, the latent mechanical representation inference model can be implemented using machine learning models, deep neural network models, image-feature fusion models, attention models, Transformer models, or other models that can infer implicit mechanical feature representations based on B-mode structural features.

[0127] Furthermore, the latent tissue biomechanical representation is a case-level intermediate feature representation generated for the current input lesion. It characterizes the overall relative mechanical state of the current lesion, the heterogeneity of its internal mechanical distribution, and the relative mechanical relationship between the lesion and adjacent reference tissue areas. The latent tissue biomechanical representation can be a feature vector, feature matrix, or feature map with spatial location information. Unlike the general correspondence rules in the prior knowledge base, the latent tissue biomechanical representation carries individualized mechanical information related to the current lesion. It should be noted that the latent tissue biomechanical representation is not a direct measurement of the actual tissue elastic modulus, strain value, or other tissue mechanical parameters. Instead, it is an implicit mechanically relevant feature representation inferred from the model based on the prior information of the current lesion's B-mode structural features and stiffness distribution, used as individualized conditional information for the subsequent construction of virtual elastography.

[0128] In this embodiment, by first acquiring the structural features of the B-mode image to be processed, and then matching the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, the potential characterization of tissue mechanics is obtained. With the help of the pre-constructed prior mapping, the implicit mechanical features of the B-mode image can be efficiently extracted, providing accurate and reliable basic data of tissue mechanics for the construction of virtual elastic imaging. This simplifies the mechanical feature extraction process and improves the accuracy and efficiency of elastic imaging reconstruction.

[0129] Based on the prior knowledge base of B-mode features and stiffness distribution, in order to automatically construct virtual elastography for the B-mode images to be processed, a lesion segmentation network is first used to identify targets in the B-mode images to be processed, obtaining lesion region images, lesion region masks, adjacent reference tissue regions, potential tissue mechanical representations, and background tissue regions. Then, a dual-branch virtual elastography construction network is constructed, which includes a lesion region mechanical reconstruction branch, a background tissue environment preservation branch, and a region constraint fusion module.

[0130] The lesion region mechanical reconstruction branch takes the lesion region image, lesion region mask, adjacent reference tissue area and potential tissue mechanical representation as input, and the elastic imaging information corresponding to the lesion region and adjacent reference tissue area as output. The potential tissue mechanical representation serves as the mechanical condition information of the current lesion and is used to guide the reconstruction of virtual elastic information corresponding to the lesion region and adjacent reference tissue area.

[0131] The Background Tissue Environment Preservation Branch takes a complete B-mode image and background tissue area as input and outputs image background information. It aims to maintain the consistency of the tissue environment, spatial structure and image background around the lesion, and reduce the generation of clinically insignificant elastic colors in the background area.

[0132] The elastography information and the image background information are input into the region constraint fusion module. The region constraint fusion module can perform fusion based on the lesion area mask, the adjacent reference tissue area mask, the background tissue area mask, the region weight, the attention mechanism or the gating mechanism, so that the construction of virtual elastography focuses on the lesion area and the adjacent reference tissue area, while maintaining the background tissue structure.

[0133] Then, based on the B-mode feature-hardness distribution prior knowledge base, a hardness distribution prior constraint is applied to the reconstruction results of the bi-branch network, so that the generated virtual elastography image conforms to the basic distribution law of real elastography in the paired training data in terms of the overall relative hardness of the lesion area, the heterogeneity of hardness distribution inside the lesion, and the relative hardness relationship between the lesion and the adjacent reference tissue area.

[0134] The network for constructing dual-branch virtual elastic imaging can be an encoder-decoder model, a U-Net structure model, a conditional generation model, a diffusion model, a conditional diffusion model, a generative adversarial network, a Transformer model, a visual attention model, or other artificial intelligence models capable of image reconstruction, and is not limited here.

[0135] Furthermore, during the training of the network for constructing the bi-branch virtual elastography, real elastography images are used as the supervision target to constrain the construction results of the virtual elastography. Training constraints may include image reconstruction loss, structural similarity loss, lesion region weighted loss, stiffness distribution prior constraint loss, background region suppression loss, perceptual loss, adversarial loss, or any combination thereof.

[0136] Among them, the hardness distribution prior constraint loss is used to constrain the consistency between the generated virtual elastic imaging image and the hardness distribution law output by the B-mode feature-hardness distribution prior knowledge base. The hardness distribution prior constraint loss may include one or more of the following: overall relative hardness consistency loss of the lesion area, hardness distribution heterogeneity consistency loss within the lesion, and relative hardness relationship consistency loss between the lesion and the adjacent reference tissue area.

[0137] Background region suppression loss is used to constrain the ineffective elastic color generation in the background region, so that the model focuses on reconstructing the elastic performance of the lesion region and the adjacent reference tissue region, reducing false color noise or clinically meaningless elastic display in the background region.

[0138] In one specific embodiment, after fusing the elastography information and the image background information to obtain the virtual elastography of the B-mode image to be processed in S308, the method further includes: evaluating the credibility of the virtual elastography based on a preset credibility evaluation index; wherein, the credibility evaluation index includes at least one of the following: the quality of the B-mode image to be processed, the confidence level of lesion segmentation, the stability of B-mode structural features, the consistency of potential tissue mechanical characterization, the degree of matching of prior knowledge base, the consistency of bi-branch reconstruction results, and the effect of background region suppression.

[0139] Among them, the preset credibility evaluation index refers to a series of quantitative indicators used to evaluate the reliability of virtual elastography image reconstruction results. It can comprehensively evaluate the construction results of virtual elastography from multiple dimensions such as image quality, lesion identification accuracy, feature extraction stability, mechanical characterization consistency, prior knowledge matching degree, reconstruction result consistency and background suppression effect, and provide a reference for clinical application.

[0140] The quality of the B-mode image to be processed refers to the comprehensive performance of the endoscopic ultrasound B-mode image in terms of sharpness, brightness, contrast, artifact level, lesion display completeness, lesion identifiability, image freeze stability, and the completeness of image acquisition information. During the construction of virtual elastography, quality control of the B-mode images to be processed is necessary. Images that are severely blurred, overexposed, underexposed, have obvious artifacts, incomplete lesion display, unidentifiable lesions, unstable image freeze, or incomplete acquisition information should be discarded or prompted for reselection to ensure the reliability of the virtual elastography construction results.

[0141] Lesion segmentation confidence refers to the reliability of the lesion segmentation network's identification results in the B-mode image to be processed. In the construction of virtual elastography, the lesion segmentation network is used to identify lesion regions, adjacent reference tissue regions, and background tissue regions in the B-mode image to be processed. Lesion segmentation confidence reflects the accuracy and reliability of the identification results and is one of the important indicators for evaluating the credibility of the virtual elastography construction results.

[0142] B-mode structural feature stability refers to the consistency and reliability of B-mode structural features extracted from the B-mode image to be processed. B-mode structural features include lesion grayscale features, lesion texture features, lesion boundary and morphological features, and contrast features between the lesion and adjacent reference tissue areas. These features are used to infer the potential tissue mechanical characterization of the current lesion. B-mode structural feature stability reflects the reliability of the feature extraction process and is one of the important indicators for evaluating the credibility of the virtual elastography construction results.

[0143] The consistency of latent tissue mechanical characterization refers to the degree of agreement between the latent tissue mechanical characterization inferred from the B-mode structural features of the B-mode image to be processed and the corresponding hardness distribution pattern in the prior knowledge base of B-mode features-hardness distribution. Late tissue mechanical characterization is used to guide the construction of virtual elastography, and its consistency reflects the reliability of the inference process, making it one of the important indicators for evaluating the credibility of the constructed virtual elastography results.

[0144] The prior knowledge base matching degree refers to the degree of matching between the structural features of the B-mode to be processed and the existing mapping relationship in the prior knowledge base of B-mode features-hardness distribution.

[0145] The consistency of bi-branch reconstruction results refers to the degree of consistency between the output results of the lesion region mechanical reconstruction branch and the background tissue environment preservation branch in the virtual elastography construction network. The lesion region mechanical reconstruction branch is used to reconstruct the elastic information of the lesion region and adjacent reference tissue areas, while the background tissue environment preservation branch is used to maintain the consistency of the surrounding tissue environment, spatial structure, and image background. The consistency of bi-branch reconstruction results reflects the reliability of the collaborative work of the bi-branch network and is one of the important indicators for evaluating the credibility of the virtual elastography construction results.

[0146] Background suppression refers to the degree to which clinically insignificant elastic color generation in the background region is suppressed during the construction of virtual elastography. In constructing virtual elastography, it is necessary to reduce false-color noise or clinically insignificant elasticity display in the background region, allowing the model to focus on reconstructing the elasticity of the lesion area and adjacent reference tissue areas. Background suppression reflects the degree to which ineffective elastic information generation in the background region is suppressed and is one of the important indicators for evaluating the reliability of the constructed virtual elastography results.

[0147] In this embodiment, after generating virtual elastography of the B-mode image to be processed, the accuracy and reliability of virtual elastography can be effectively verified by using a preset credibility evaluation index that covers at least one dimension, such as the quality of the B-mode image to be processed, the confidence level of lesion segmentation, and the stability of B-mode structural features. This provides more credible imaging support for clinical diagnosis and tissue mechanical analysis, and improves the quality of the virtual elastography construction results and its clinical application value.

[0148] For a clear and intuitive demonstration of the virtual elastic imaging process, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of an embodiment of the virtual elastography construction system provided by the present invention. First, S1 acquires and preprocesses image data to provide qualified basic data for subsequent steps. Next, S2 performs lesion region segmentation and region constraint construction to accurately locate the lesion range and establish relevant constraints. Then, S3 extracts B-mode structural features, and then S4 constructs a priori knowledge base of B-mode features-stiffness distribution to provide prior support for subsequent tissue mechanics analysis. Then, S5 is carried out to infer potential tissue mechanics characteristics and mine the mechanical features of the tissue. Subsequently, the construction of the dual-branch virtual elastography is carried out in S6, which also combines the training data of "B-mode image + real elastic image" to optimize the reconstruction process. Finally, S7 completes the output of virtual elastography results and evaluates the credibility of the results to obtain elastography results with reference value.

[0149] Further, please refer to Figure 5 , Figure 5 This is a flowchart illustrating an embodiment of the construction of a B-mode feature-hardness distribution prior knowledge base provided by the present invention. The B-mode image is a conventional ultrasound structural image that presents the morphological and structural information of tissues; the true elastic image is an ultrasound image reflecting the hardness distribution of tissues, demonstrating the hardness differences between different tissues. The dashed line defines the effective range of the elastic imaging. Specifically, the analysis process for these two types of images is as follows: first, image quality control is performed to ensure image usability; then, the lesion area is labeled or identified; subsequently, the two types of images are paired and spatially registered to ensure their spatial positions correspond; finally, a B-mode feature-hardness distribution prior knowledge base is constructed to support relevant medical diagnosis or research.

[0150] Further, please refer to Figure 6 , Figure 6This is a schematic diagram of the structure of an embodiment of the dual-branch virtual elastography network provided by the present invention. First, a B-mode ultrasound image is input and enters the core processing unit. On the one hand, the global feature encoder extracts global features of the image and combines them with the lesion area conditions. At the same time, the A branch completes the mechanical reconstruction of the lesion area, and the B branch completes the task of maintaining the background tissue environment. These features and branch results enter the feature fusion and region constraint stage, and are then sent to the image reconstruction module for image generation. Meanwhile, the image reconstruction module is subject to various loss functions, including reconstruction loss, structural similarity loss, lesion area weighted loss, hardness distribution prior constraint loss, and background area suppression loss, in order to optimize the generation effect. Finally, a virtual elastography image is output, realizing the conversion from a conventional ultrasound image to an elastography image that can reflect tissue hardness information, and assisting in the evaluation of related lesions.

[0151] To facilitate better implementation of the virtual elastic imaging construction method provided by this invention, this invention also provides a system based on the above-described virtual elastic imaging construction method. The meanings of the terms used are the same as in the virtual elastic imaging construction method described above, and specific implementation details can be found in the descriptions of the method embodiments.

[0152] Please see Figure 7 , Figure 7 This is a schematic diagram of an embodiment of the virtual elastic imaging construction system provided by the present invention, wherein the virtual elastic imaging construction system 700 may include: The sample acquisition module 701 is used to acquire a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elastic imaging image. The B-mode image processing module 702 is used to divide the B-mode image into regions to obtain a first region set of the B-mode image, and to extract features from each region of the first region set to obtain the B-mode structural features of each region. The real elastic imaging image processing module 703 is used to determine the second region set of the real elastic imaging image corresponding to the B mode image based on the division of the first region set, and to extract the hardness features of each region in the second region set to obtain the relative hardness distribution features of each region. The knowledge base construction module 704 is used to perform feature mapping on the B-mode structural features and relative hardness distribution features of the same type of lesions, and to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. The implicit mechanical feature extraction module 705 is used to extract implicit mechanical features from the B-mode image to be processed based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, so as to obtain the tissue mechanical potential characterization of the B-mode image to be processed. The elastography information generation module 706 is used to generate elastography information of the B-mode image to be processed based on the potential tissue mechanics characterization and the lesion-related region of the B-mode image to be processed. The image background information generation module 707 is used to generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background area of ​​the B-mode image to be processed. The virtual elastic imaging module 708 is used to fuse elastic imaging information and image background information to obtain virtual elastic imaging of the B-mode image to be processed.

[0153] The present invention also provides an electronic device, such as... Figure 8 As shown, Figure 8 This is a schematic diagram of an embodiment of the electronic device provided by the present invention, specifically: The electronic device may include components such as a processor 801 with one or more processing cores, a memory 802 with one or more computer-readable storage media, a power supply 803, and an input unit 804. Those skilled in the art will understand that... Figure 8 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 801 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 802, and by calling data stored in the memory 802, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 801 may include one or more processing cores; preferably, the processor 801 may integrate an application processor and a modem processor. The application processor mainly handles the operation of the storage medium, user interface, and application programs, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 801.

[0154] The memory 802 can be used to store software programs and modules. The processor 801 executes various functional applications and data processing by running the software programs and modules stored in the memory 802. The memory 802 may mainly include a program storage area and a data storage area. The program storage area may store application programs required for operating the storage medium and at least one function (such as sound playback function, image playback function, etc.); the data storage area may store data created according to the use of the electronic device. In addition, the memory 802 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 802 may also include a memory controller to provide the processor 801 with access to the memory 802.

[0155] The electronic device also includes a power supply 803 that supplies power to various components. Preferably, the power supply 803 can be logically connected to the processor 801 via a power management storage medium, thereby enabling functions such as charging, discharging, and power consumption management through the power management storage medium. The power supply 803 may also include one or more DC or AC power supplies, recharge storage media, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0156] The electronic device may also include an input unit 804, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0157] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 801 in the electronic device loads the executable files corresponding to the processes of one or more application programs into the memory 802 according to the following instructions, and the processor 801 runs the application programs stored in the memory 802 to realize various functions, as follows: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain the first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. Feature mapping is performed on the B-mode structural features and relative hardness distribution features of the same type of lesions to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features and hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related regions of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate the image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed. The elastic imaging information and the image background information are fused to obtain a virtual elastic imaging of the B-mode image to be processed.

[0158] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0159] Therefore, the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the steps in any of the virtual elastic imaging construction methods provided by the present invention. For example, the computer program, when loaded by a processor, can execute the following steps: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain the first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. Feature mapping is performed on the B-mode structural features and relative hardness distribution features of the same type of lesions to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features and hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related regions of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate the image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed. The elastic imaging information and the image background information are fused to obtain a virtual elastic imaging of the B-mode image to be processed.

[0160] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0161] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0162] Since the computer program stored in the computer-readable storage medium can execute the steps in any of the virtual elastic imaging construction methods provided by the present invention, the beneficial effects that any of the virtual elastic imaging construction methods provided by the present invention can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0163] The present invention has provided a detailed description of a method, system, and electronic device for constructing virtual elastic imaging. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for constructing virtual elastic imaging, characterized in that, include: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain a first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. For the same type of lesion, feature mapping is performed on the B-mode structural features and the relative hardness distribution features to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related region of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed; The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

2. The method for constructing virtual elastic imaging according to claim 1, characterized in that, The acquisition of the sample data set includes: Filter the initial sample data pairs from the same case, the same examination procedure, the same lesion, or the same image scene; By removing outlier data pairs, incomplete data pairs, and artifact data pairs from the initial sample data pair set, a first sample data pair set is obtained. The first set of sample data pairs is subjected to standardization preprocessing and spatial registration to obtain the set of sample data pairs. The standardized preprocessing includes at least one of image cropping, size normalization, grayscale normalization, brightness and contrast correction, noise suppression, removal of irrelevant text or borders, image orientation unification, color space conversion, and image format conversion. The spatial registration includes at least one of deep learning, affine transformation, or non-rigid deformation.

3. The method for constructing virtual elastic imaging according to claim 1, characterized in that, The first region set includes the lesion region, the adjacent reference tissue region, and the background tissue region; the process of dividing the B-mode image into regions to obtain the first region set of the B-mode image includes: Target recognition is performed on the B-mode image to obtain a lesion region mask, and the lesion region is determined based on the lesion region mask; The region in the B-mode image that is at a preset distance from the lesion region is defined as the adjacent reference tissue region; The remaining area in the B-mode image, excluding the lesion area and the adjacent reference tissue area, is defined as the background tissue area.

4. The method for constructing virtual elastic imaging according to claim 3, characterized in that, The structural features of the B-mode include grayscale features of lesions, texture features of lesions, boundary and morphological features of lesions, and contrast features. The step of extracting features from each region of the first region set to obtain the B-mode structural features of each region includes: Gray-scale statistical analysis and texture feature calculation are performed on the lesion area to obtain the gray-scale features and texture features of the lesion; Edge detection and morphological analysis are performed based on the lesion region mask and the B-mode image to extract the lesion boundary and morphological features; Gray-scale statistical analysis and texture feature calculation are performed on the adjacent reference tissue area to obtain the gray-scale features and texture features of the adjacent reference tissue area. The gray-scale features and texture features are then compared with the gray-scale features and texture features of the lesion to obtain the comparison features.

5. The method for constructing virtual elastic imaging according to claim 4, characterized in that, The relative hardness distribution features include the overall relative hardness of the lesion area, the heterogeneity of hardness distribution within the lesion, and the relative hardness difference between the lesion and the adjacent reference tissue area; based on the division of the first region set, a second region set of the true elastography image corresponding to the B-mode image is determined, and hardness features are extracted from each region of the second region set to obtain the relative hardness distribution features of each region, including: Based on the division of the lesion area and the adjacent reference tissue area, the true lesion area and the true adjacent reference tissue area of ​​the true elastography image are determined respectively. The overall relative hardness of the lesion region and the heterogeneity of hardness distribution within the lesion are obtained from the actual lesion region. The hardness characteristics of the actual lesion area and the actual adjacent reference tissue area are compared and analyzed to obtain the relative hardness difference between the lesion and the adjacent reference tissue area.

6. The method for constructing virtual elastic imaging according to claim 1 or 5, characterized in that, The method involves feature mapping of the B-mode structural features and the relative hardness distribution features of the same type of lesions to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion, including: The sample data sets are clustered and grouped according to the lesion type to obtain sample data pairs for each type of lesion. Based on the correspondence of the sample data pairs, the B-mode structural features and the relative hardness distribution features are spatially registered and associated to obtain the B-mode feature-hardness distribution mapping relationship for each sample data pair. Based on the sample data, the B-mode feature-hardness distribution mapping relationship within the group is statistically mapped to obtain the prior knowledge base of B-mode feature-hardness distribution for each type of lesion.

7. The method for constructing virtual elastic imaging according to claim 1, characterized in that, The implicit mechanical feature extraction of the B-mode image to be processed, based on the mapping relationship in the prior knowledge base of B-mode features and hardness distribution, yields the potential tissue mechanical characterization of the B-mode image to be processed, including: Obtain the structural features of the B-mode image to be processed; The mapping relationship between the structural features of the B-mode to be processed and the prior knowledge base of the B-mode features-hardness distribution is matched to obtain the potential tissue mechanical characterization of the B-mode image to be processed.

8. The method for constructing virtual elastic imaging according to claim 1, characterized in that, After fusing the elastic imaging information and the image background information to obtain the virtual elastic imaging of the B-mode image to be processed, the method further includes: The credibility of the virtual elastic imaging is evaluated based on a preset credibility evaluation index. The credibility evaluation index includes at least one of the following: the quality of the B-mode image to be processed, the confidence level of lesion segmentation, the stability of B-mode structural features, the consistency of potential tissue mechanical representation, the degree of matching of prior knowledge base, the consistency of bi-branch reconstruction results, and the effect of background region suppression.

9. A system for constructing virtual elastic imaging, characterized in that, include: The sample acquisition module is used to acquire a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image. The B-mode image processing module is used to divide the B-mode image into regions to obtain a first region set of the B-mode image, and to extract features from each region of the first region set to obtain the B-mode structural features of each region. The real elastic imaging image processing module is used to determine the second region set of the real elastic imaging image corresponding to the B mode image based on the division of the first region set, and to extract the hardness features of each region in the second region set to obtain the relative hardness distribution features of each region. The knowledge base construction module is used to perform feature mapping on the B-mode structural features and the relative hardness distribution features of the same type of lesions, and to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. The implicit mechanical feature extraction module is used to extract implicit mechanical features from the B-mode image to be processed based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, so as to obtain the potential tissue mechanical characterization of the B-mode image to be processed. An elastography information generation module is used to generate elastography information of the B-mode image to be processed based on the potential tissue biomechanical characterization and the lesion-related region of the B-mode image to be processed. An image background information generation module is used to generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background area of ​​the B-mode image to be processed. A virtual elastic imaging module is used to fuse the elastic imaging information and the image background information to obtain a virtual elastic imaging of the B-mode image to be processed.

10. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the following steps: Obtain a set of sample data pairs, wherein each sample data pair includes a B-mode image and its corresponding real elasticity imaging image; The B-mode image is divided into regions to obtain a first region set of the B-mode image, and features are extracted from each region of the first region set to obtain the B-mode structural features of each region. Based on the division of the first region set, the second region set of the real elastic imaging image corresponding to the B mode image is determined, and the hardness features of each region in the second region set are extracted to obtain the relative hardness distribution features of each region. For the same type of lesion, feature mapping is performed on the B-mode structural features and the relative hardness distribution features to establish a prior knowledge base of B-mode features-hardness distribution for each type of lesion. Based on the mapping relationship in the prior knowledge base of B-mode features-hardness distribution, implicit mechanical features are extracted from the B-mode image to be processed to obtain the potential tissue mechanical characterization of the B-mode image to be processed. Based on the potential biomechanical characterization of the tissue and the lesion-related region of the B-mode image to be processed, elastography information of the B-mode image to be processed is generated. Generate image background information of the B-mode image to be processed based on the B-mode image to be processed and the background region of the B-mode image to be processed; The elastic imaging information and the image background information are fused to obtain the virtual elastic imaging of the B-mode image to be processed.

Images