Image generation method, apparatus, device, storage medium, and program product

Abstract

The application relates to an image generation method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring multiple T1 weighted images and a CINE image of a to-be-tested tissue; performing image generation processing according to the multiple T1 weighted images, the CINE image and a preset neural network to generate a virtual LGE image corresponding to the to-be-tested tissue; in the virtual LGE image, the signal contrast of an abnormal region in the to-be-tested tissue is higher than that of other regions; the neural network is obtained by training based on multiple training sample sets and a gold standard corresponding to each training sample set; each training sample set comprises multiple training T1 weighted images, a training T1 mapping image and a training CINE image, and the gold standard comprises a gold standard LGE image. The method can avoid injecting gadolinium contrast agent into a subject, and can also ensure the smooth execution of the imaging process or the examination process of LGE delayed gadolinium enhancement imaging.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image generation method, apparatus, device, storage medium, and program product. Background Technology

[0002] Magnetic resonance imaging (MRI) is widely used in cardiac imaging due to its excellent soft tissue contrast and multi-angle imaging capabilities. MRI typically involves T1-weighted images, which are acquired using short repetition times (TR) and echo times (TE) sequences, primarily highlighting longitudinal relaxation differences in tissues. While T1-weighted images can quantitatively observe abnormal signals or regions in tissues to some extent, the observational results are not ideal.

[0003] Currently, in order to better observe abnormal signals or areas in tissues, delayed gadolinium-enhanced imaging (LGE) is generally used, in which gadolinium contrast agent is injected into the tissue of the examinee to obtain LGE images. The enhanced areas in the LGE images are the abnormal signals or areas in the tissue. In this way, abnormal signals or areas in the tissue can be clearly observed through LGE images.

[0004] However, some patients are allergic to the injected gadolinium contrast agent during LGE delayed gadolinium contrast imaging, which can affect the smooth execution of the imaging or examination process. Summary of the Invention

[0005] Therefore, it is necessary to provide an image generation method, apparatus, device, storage medium, and program product that can avoid injecting gadolinium contrast agent into the subject and ensure the smooth execution of the imaging or examination process of LGE delayed gadolinium enhancement imaging, in order to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides an image generation method, the method comprising:

[0007] Acquire multiple T1-weighted images and CINE images of the tissue to be tested;

[0008] Image generation processing is performed based on multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image corresponding to the tissue under test.

[0009] In the virtual LGE image, the signal contrast of the abnormal area in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images. The gold standard includes the gold standard LGE image.

[0010] In one embodiment, the above-mentioned image generation process, which generates a virtual LGE image corresponding to the tissue under test based on multiple T1-weighted images, CINE images, and a preset neural network, includes:

[0011] Determine the corresponding T1 mapping image based on multiple T1 weighted images;

[0012] Based on multiple T1-weighted images, T1 mapping images, and CINE images, the first fusion feature corresponding to the tissue under test is determined;

[0013] The first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image.

[0014] In one embodiment, determining the first fusion feature corresponding to the tissue under test based on multiple T1-weighted images, T1 mapping images, and CINE images includes:

[0015] Feature extraction processing is performed on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference relationship between abnormal areas and other areas in the tissue under test.

[0016] Multiple T1-weighted images, T1 mapping images, CINE images, and the first feature map are fused to determine the first fused feature.

[0017] In one embodiment, the above-mentioned feature extraction processing of multiple T1-weighted images and T1 mapping images to determine the first feature map corresponding to the tissue to be tested includes:

[0018] Feature extraction is performed on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue under test.

[0019] Feature extraction processing is performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test;

[0020] The first and second sub-feature maps are fused to determine the first feature map.

[0021] In one embodiment, the above-mentioned image generation process based on multiple T1-weighted images, CINE images, and a preset neural network to generate a virtual LGE image corresponding to the tissue under test further includes:

[0022] Image generation processing is performed using multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0023] In one embodiment, the above-mentioned image generation process based on multiple T1-weighted images, CINE images, and a preset neural network to generate an abnormal region segmentation mask corresponding to the tissue under test includes:

[0024] Based on multiple T1-weighted images, determine the first feature map corresponding to the tissue to be tested;

[0025] Feature extraction processing is performed on the CINE image to determine the second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during the motion process;

[0026] The first feature map and the second feature map are fused to determine the second fused feature.

[0027] The second fusion feature is input into the neural network for image segmentation processing to determine the abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes the abnormal region in the tissue under test.

[0028] In one embodiment, the gold standard further includes a gold standard anomaly region segmentation mask, and the training method of the neural network includes:

[0029] Based on each training sample set and the initial neural network, determine the predicted LGE image and the predicted anomaly region segmentation mask corresponding to each training image subset;

[0030] The initial neural network is trained based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, and the neural network is determined.

[0031] In one embodiment, the above-mentioned training of the initial neural network based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training image subset, and determining the neural network, includes:

[0032] Calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each training sample set;

[0033] Calculate the second loss between the predicted anomaly region segmentation mask and the corresponding gold standard anomaly region segmentation mask for each training sample set;

[0034] The initial neural network is trained based on the first loss and the second loss to determine the neural network.

[0035] Secondly, this application also provides an image generation apparatus, the apparatus comprising:

[0036] The image acquisition module is used to acquire multiple T1-weighted images and CINE images of the tissue under test;

[0037] The image generation module is used to generate a virtual LGE image corresponding to the tissue under test by performing image generation processing based on multiple T1-weighted images, CINE images, and a preset neural network. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes the gold standard LGE image.

[0038] Thirdly, this application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0039] Acquire multiple T1-weighted images and CINE images of the tissue to be tested;

[0040] Image generation processing is performed based on multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image corresponding to the tissue under test.

[0041] In the virtual LGE image, the signal contrast of the abnormal area in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images. The gold standard includes the gold standard LGE image.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0043] Acquire multiple T1-weighted images and CINE images of the tissue to be tested;

[0044] Image generation processing is performed based on multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image corresponding to the tissue under test.

[0045] In the virtual LGE image, the signal contrast of the abnormal area in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images. The gold standard includes the gold standard LGE image.

[0046] Fifthly, this application also provides a computer program product, comprising a computer program that, when executed by a processor, performs the following steps:

[0047] Acquire multiple T1-weighted images and CINE images of the tissue to be tested;

[0048] Image generation processing is performed based on multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image corresponding to the tissue under test.

[0049] In the virtual LGE image, the signal contrast of the abnormal area in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images. The gold standard includes the gold standard LGE image.

[0050] The aforementioned image generation method, apparatus, device, storage medium, and program product generate a virtual LGE image corresponding to the tissue under test by processing multiple acquired T1-weighted images and CINE images with a preset neural network. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and corresponding gold standards. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes a gold standard LGE image. In this method, since the neural network for generating virtual LGE images can be trained using different training sample sets and gold standard LGE images, a virtual LGE image of the tissue under test can be generated by combining the T1-weighted image and CINE image of the tissue under test with this neural network, without the need to inject gadolinium contrast agent into the subject to obtain the LGE image. This avoids the possibility of the subject being allergic to gadolinium contrast agent, improves the user examination experience, and ensures the smooth execution of LGE delayed gadolinium enhancement imaging of the tissue under test and subsequent examination procedures. Attached Figure Description

[0051] Figure 1This is an internal structural diagram of a computer device in one embodiment;

[0052] Figure 2 This is a flowchart illustrating an image generation method in one embodiment;

[0053] Figure 3 This is a flowchart illustrating the image generation method in another embodiment;

[0054] Figure 4 This is a flowchart illustrating the image generation method in another embodiment;

[0055] Figure 5 This is a flowchart illustrating the image generation method in another embodiment;

[0056] Figure 6 This is a flowchart illustrating the image generation method in another embodiment;

[0057] Figure 7 This is a schematic diagram of the flow framework of an image generation method in another embodiment;

[0058] Figure 8 This is a structural block diagram of an image generation device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] Magnetic resonance imaging (MRI) is widely used in cardiac imaging due to its excellent soft tissue contrast and multi-angle imaging capabilities. T1 mapping images, in particular, can reflect the longitudinal relaxation time of tissues. Currently, to understand the different delayed enhancement patterns in various cardiac diseases, delayed gadolinium contrast-enhanced imaging (LGE) requiring injection (i.e., images obtained by injecting gadolinium contrast agent) is commonly used clinically. The mechanism of LGE is mainly based on the following two points: 1. Gadolinium contrast agent is an extracellular contrast agent and cannot pass through normal cell membranes; 2. Normal cardiomyocytes are densely packed, occupying most of the space in the myocardial tissue, while the intercellular matrix occupies a very small proportion, thus gadolinium contrast agent is sparsely distributed in normal myocardium. In acute myocardial injury (such as acute myocardial infarction, acute myocarditis), the myocardial cell membrane ruptures, and gadolinium contrast agent diffuses into the cells through the damaged cell membrane. Combined with intercellular edema, this leads to an increase in the extracellular space, resulting in an increased local concentration of gadolinium contrast agent. In chronic myocardial injury (such as chronic myocardial infarction and non-ischemic cardiomyopathy), myocardial cells are replaced by fibrous scar tissue. The interstitial spaces between collagen fibers are significantly larger than those between normal myocardial cells, resulting in gadolinium contrast agent concentration in the scar area. It is evident that current technology has the following problems: firstly, some patients are allergic to gadolinium contrast agents, or their metabolism is burdened; secondly, delayed gadolinium enhancement sequences increase the scanning burden on both patients and doctors, affecting the smooth execution of the imaging or examination process. Based on this, embodiments of this application provide an image generation method, apparatus, device, storage medium, and program product that can solve the above-mentioned technical problems.

[0061] The image generation method provided in this application embodiment can be applied to computer devices, which can be terminals or servers. Taking a terminal as an example, its internal structure diagram can be as follows: Figure 1 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an image generation method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0062] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0063] In one embodiment, such as Figure 2 As shown, an image generation method is provided, which can be applied to... Figure 1 Taking a computer device as an example, the method may include the following steps:

[0064] S202, acquire multiple T1-weighted images and CINE images of the tissue to be tested.

[0065] In this step, the tissue to be tested can be the heart, the myocardium within the heart, a normal area of ​​the myocardium, or a damaged area of ​​the myocardium, etc. Of course, the tissue to be tested can also be other tissues, such as the brain, the kidneys in the abdomen, the liver, etc.

[0066] The T1-weighted image can be referred to as the T1 Weight image. The T1-weighted image can be obtained by scanning the tissue to be tested using a scanning sequence and calculating the T1-weighted image from multiple echo data collected. The multiple T1-weighted images here can be T1-weighted images obtained by calculating different echo data in the above manner.

[0067] The CINE images here can also be called movie images, which can include images of different motion states during the heartbeat.

[0068] It should be noted that the multiple T1-weighted images and CINE images here were all acquired from the same site of the same tissue being tested, and were all acquired without the injection of gadolinium contrast agent.

[0069] S204 generates a virtual LGE image corresponding to the tissue under test by performing image generation processing based on multiple T1-weighted images, CINE images, and a preset neural network.

[0070] The network type and architecture of the neural network can be set according to the actual situation, and no specific limitation is made here. In addition, the neural network can be a neural network composed of a single-level neural network, a neural network composed of multiple cascaded sub-networks, or a neural network composed of neural networks with different functions in parallel. For example, it can be a neural network composed of an image generation network and an image segmentation network in parallel, or other situations.

[0071] The neural network here is primarily used to generate virtual LGE images, which can be called delayed gadolinium-enhanced magnetic resonance images. Virtual LGE images are those generated directly by the network without the need for gadolinium contrast agent injection. In these generated virtual LGE images, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. This simulates a real LGE image, facilitating the rapid and accurate identification of abnormal areas in the tissue under test, which is beneficial for subsequent image analysis.

[0072] Furthermore, the neural network here is generally pre-trained, which can be based on multiple training sample sets and a corresponding gold standard for each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes a gold standard LGE image. The multiple T1-weighted images, training T1 mapping images, and training CINE images in each training sample set are all images of the same sample tissue. The training sample sets can be training images collected from different sample tissues, or they can be training images collected from the same sample group. Additionally, each training sample set corresponds to only one gold standard LGE image, which can be an image formed after injecting gadolinium contrast agent into the sample tissue. The T1 mapping image, also known as high-resolution longitudinal relaxation time quantitative imaging, can be obtained by converting the training T1-weighted images or by directly calculating from the scan data.

[0073] After training the neural network, multiple T1-weighted images and CINE images of the tissue under test can be directly input into the neural network for LGE image generation. Alternatively, multiple T1-weighted images and CINE images of the tissue under test can be preprocessed (this preprocessing may include image registration, etc.) before being input into the neural network for LGE image generation. Or, multiple T1-weighted images and CINE images of the tissue under test can be combined with other types of images and input into the neural network for LGE image generation. In short, the final LGE image, denoted as virtual LGE image, can be obtained through image generation processing via the neural network.

[0074] Taking myocardium as the tissue to be tested as an example, this embodiment uses T1-weighted images and CINE images when generating virtual LGE images. T1-weighted images can provide quantitative theoretical support for the process of generating virtual LGE images of the tissue to be tested, while CINE images can reflect the pattern of myocardial motion in the time dimension, so that the generated virtual LGE images can conform to the actual myocardial motion. In this way, virtual LGE images can be accurately and quantitatively generated through neural networks, thereby improving the accuracy of the obtained virtual LGE images.

[0075] This virtual LGE image can be generated directly by processing various types of images through a neural network without injecting gadolinium contrast agent into the subject's tissue. Therefore, it can reduce the risk of subject allergy to gadolinium contrast agent, ensure the smooth execution of the delayed gadolinium-enhanced magnetic resonance imaging process, and provide an accurate data foundation for subsequent analysis processes that require LGE images.

[0076] In the aforementioned image generation method, multiple T1-weighted images and CINE images, along with a pre-set neural network, are used to generate a virtual LGE image corresponding to the tissue under test. In this virtual LGE image, the signal contrast of abnormal areas within the tissue is higher than that of other areas. The neural network is trained based on multiple training sample sets and corresponding gold standards. Each training sample set includes multiple training T1-weighted images, training T1-mapping images, and training CINE images. The gold standard includes a gold standard LGE image. This method allows the neural network for generating virtual LGE images to be trained using different training sample sets and gold standard LGE images. This means that the virtual LGE image of the tissue under test can be generated by combining the T1-weighted image and CINE image with the neural network, eliminating the need to inject gadolinium contrast agent into the patient to obtain the LGE image. This avoids the possibility of patient allergies to gadolinium contrast agents, improves the user's examination experience, and ensures the smooth execution of delayed gadolinium-enhanced LGE imaging of the tissue under test and subsequent examination procedures.

[0077] The above embodiments mentioned that virtual LGE images can be generated by combining multiple T1-weighted images, CINE images, and other types of images through a neural network. The following embodiments illustrate this process.

[0078] In another embodiment, another image generation method is provided, based on the above embodiments, such as... Figure 3 As shown, the above S204 may include the following steps:

[0079] S302, determine the corresponding T1 mapping image based on multiple T1 weighted images.

[0080] In this step, after obtaining multiple T1-weighted images, which are generally of the same size, we can perform mean or fitting processing on these multiple T1-weighted images to obtain the corresponding T1 mapping image.

[0081] Of course, other methods can also be used to obtain the corresponding T1 mapping image from multiple T1 weighted images, which is not specifically limited here.

[0082] S304. Based on multiple T1-weighted images, T1 mapping images, and CINE images, determine the first fusion feature corresponding to the tissue under test.

[0083] In this step, after obtaining multiple T1-weighted images, T1 mapping images, and CINE images, these different types of images can be fused to obtain the first fusion feature of the tissue under test. Alternatively, features can be extracted from these different types of images, and the feature maps of different types of images can be fused to obtain the first fusion feature. Another approach is to extract features from these different types of images, and then fuse both the feature maps of different types of images and the images themselves to obtain the first fusion feature. Other methods may also be used, as long as the first fusion feature can be obtained.

[0084] In this step, subsequent image generation is performed by combining different types of images or features. This combination of information results in a richer and more accurate virtual LGE image.

[0085] S306, the first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image.

[0086] In this step, after obtaining the first fusion feature of the tissue to be tested, the first fusion feature can be input into the neural network for image generation processing. Alternatively, other images or features can be input into the neural network to assist in image generation processing, and finally, a virtual LGE image is obtained.

[0087] In this embodiment, a T1 mapping image is obtained from multiple T1-weighted images, and a fusion feature of the tissue under test is obtained by combining it with a CINE image. This fusion feature is then input into a neural network to obtain a virtual LGE image. This method of generating a virtual LGE image by combining multiple image types enhances the richness of the information input to the neural network, thereby further improving the accuracy of the final generated virtual LGE image. Furthermore, since the T1 mapping image can be calculated from multiple T1-weighted images, multiple scans of the subject are avoided, shortening the scanning time and improving the examinee's experience.

[0088] The above embodiments mentioned that feature extraction can be performed on various types of images to obtain the content of the fused image. The following embodiments illustrate one implementation of this process.

[0089] In another embodiment, another image generation method is provided, based on the above embodiments, such as... Figure 4 As shown, the above S304 may include the following steps:

[0090] S402, perform feature extraction processing on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference between abnormal areas and other areas in the tissue under test.

[0091] In this step, feature extraction can be performed on multiple T1-weighted images to obtain the first feature map of the tissue under test; or feature extraction can be performed on the T1 mapping image to obtain the first feature map of the tissue under test.

[0092] Alternatively, feature extraction can be performed on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue under test; feature extraction can be performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test; and the first and second sub-feature maps can be fused to determine the first feature map. That is, multiple T1-weighted images can be fused and then feature extracted to obtain the first sub-feature map, or features can be extracted from multiple T1-weighted images as a whole to obtain the first sub-feature map; then feature extraction can be performed on the T1 mapping image to obtain the second sub-feature map, and finally the first and second sub-feature maps can be fused to obtain the first feature map of the tissue under test. Here, feature extraction is performed separately on the T1-weighted image and the T1 mapping image, followed by feature fusion. This ensures that the final first feature map contains more and richer information, thus providing richer and more diverse feature information for subsequent image generation.

[0093] The above-mentioned feature extraction process for multiple T1-weighted images and / or T1 mapping images can be performed using neural networks or feature extraction algorithms, etc., without specific limitations here.

[0094] Since the T1 mapping image can be calculated from the T1 weighted image, the image information of the T1 weighted image and the T1 mapping image is relatively similar. Therefore, the difference between abnormal areas and other areas in the tissue under test can be reflected by the image information of the two or the feature map after feature extraction. This difference can be the difference between the image value of the abnormal area and the image value of other areas in the tissue under test on the feature map.

[0095] S404, perform fusion processing on multiple T1 weighted images, T1 mapping images, CINE images and the first feature map to determine the first fusion feature.

[0096] In this step, after obtaining the first feature map, the first feature map can be fused with three other types of original images (including multiple T1-weighted images, T1 mapping images, and CINE images) to obtain the first fused features of the tissue to be tested. By combining the feature map with the original images, original image information and feature information can be provided for subsequent image generation, ensuring both data richness and quantitative accuracy of the generated images.

[0097] In this embodiment, the feature map obtained by extracting features from the T1-weighted image and / or the T1 mapping image is fused with other original images of various types to obtain fused features. This ensures both data richness and quantitative accuracy of the generated image.

[0098] The above embodiments mention that the neural network is mainly used to generate virtual LGE images. Of course, the neural network can also be used to generate other types of images. The following embodiments illustrate one implementation of this process.

[0099] In another embodiment, another image generation method is provided. Based on the above embodiments, S204 may include the following steps:

[0100] Step A: Based on multiple T1-weighted images, CINE images, and a preset neural network, image generation processing is performed to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0101] In this step, multiple T1-weighted images and CINE images can be input into a neural network for image generation and segmentation, or multiple T1-weighted images and CINE images can be combined with other images and input into a neural network for image generation and segmentation, or multiple T1-weighted images and CINE images can be processed and then input into a neural network for image generation and segmentation, or other cases may exist.

[0102] By performing image generation processing on the LGE image of the tissue under test within a neural network, a virtual LGE image of the tissue under test can be obtained. Simultaneously, by performing image segmentation processing on abnormal regions within the tissue under test within a neural network, a segmentation mask for the abnormal regions within the tissue under test can be obtained.

[0103] Specifically, in the image segmentation process, as an optional embodiment, please refer to... Figure 5 The flowchart shown illustrates that the process of generating the abnormal region segmentation mask corresponding to the tissue under test in this step may include the following steps:

[0104] S502, based on multiple T1-weighted images, determine the first feature map corresponding to the tissue to be tested.

[0105] In this step, a corresponding T1 mapping image can be obtained by calculating multiple T1 weighted images. Then, feature extraction processing is performed on the multiple T1 weighted images to determine the first sub-feature map corresponding to the tissue under test. Feature extraction processing is also performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test. Finally, the first sub-feature map and the second sub-feature map are fused to determine the first feature map of the tissue under test.

[0106] S504, Perform feature extraction processing on the CINE image to determine the second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during the motion process.

[0107] In this step, neural networks or feature extraction algorithms can be used to extract features from the CINE image to obtain a feature map corresponding to the tissue under test in the CINE image, denoted as the second feature map. This second feature map can include abnormal information and contour information of the tissue under test during its movement. The abnormal information can be information about abnormal regions appearing in the tissue under test, such as the location and frequency of these regions. The contour information can include the contour information of the tissue under test, the contour information of abnormal regions within the tissue under test, and the contour information of normal regions within the tissue under test.

[0108] In addition, taking the myocardium in the heart as an example, the second feature map extracted here can be a feature map that includes high-dimensional myocardial motion abnormality information / high-dimensional myocardial abnormality information, while the first feature map mentioned above can be a low-dimensional myocardial feature map.

[0109] S506, perform fusion processing on the first feature map and the second feature map to determine the second fused feature.

[0110] In this step, after obtaining the first feature map and the second feature map, the two feature maps can be fused to obtain a fused feature map, which is denoted as the second fused feature.

[0111] S508, the second fusion feature is input into the neural network for image segmentation processing to determine the abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes the abnormal region in the tissue under test.

[0112] In this step, after obtaining the second fusion feature, the second fusion feature can be input into the neural network to perform abnormal region segmentation processing in the tissue under test, and obtain a segmentation mask of the abnormal region in the tissue under test, that is, the abnormal region segmentation mask includes the abnormal region in the tissue under test.

[0113] During segmentation, for example, the tissue to be tested can be segmented in the neural network first, and then the abnormal regions in the segmented tissue can be segmented. Alternatively, the abnormal regions in the tissue to be tested can be segmented directly.

[0114] Taking the myocardium in the heart as an example, the abnormal area here can be an abnormal area in the myocardium.

[0115] In this embodiment, a virtual LGE image and an abnormal region segmentation mask of the tissue under test are obtained by generating and segmenting images using multiple T1-weighted images, CINE images, and a neural network. This allows for the rapid acquisition of different image results through a single network for both generated and segmented images. Furthermore, by fusing multiple feature maps of different dimensions and inputting them into the neural network for segmentation to obtain the abnormal region segmentation mask, the segmentation process is refined, and the combined information is more abundant, thus improving the accuracy of the final abnormal region segmentation mask.

[0116] The above embodiments mention that the neural network can perform image generation and image segmentation processing simultaneously. It also mentions that the neural network can be trained, and the gold standard for training includes the gold standard LGE image. In addition, the gold standard can also include the gold standard abnormal region segmentation mask. That is to say, the neural network can be trained based on these two gold standards. The following embodiments illustrate the training method of the neural network.

[0117] In another embodiment, another image generation method is provided, based on the above embodiments, such as... Figure 6 As shown, the training method for the above neural network may include the following steps:

[0118] S602, based on each training sample set and the initial neural network, determine the predicted LGE image and the predicted abnormal region segmentation mask corresponding to each training image subset.

[0119] In this step, we can first obtain each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images, and training CINE images. Each training sample set is a training image obtained for the same sample organization.

[0120] Then, multiple training T1 weighted images, training T1 mapping images, and training CINE images from each training sample set can be registered to obtain each registered training sample set. After that, each registered training sample set is input into the initial neural network. In the initial neural network, feature extraction, image fusion, image generation, and image segmentation can be performed on the images in the registered training sample sets. Finally, the virtual LGE image and anomaly region segmentation mask corresponding to each training sample set can be output.

[0121] Here, the virtual LGE image and the anomaly region segmentation mask output by the initial neural network can be denoted as the predicted LGE image and the predicted anomaly region segmentation mask, respectively.

[0122] S604. Based on the predicted LGE image, the gold standard LGE image, the predicted abnormal region segmentation mask, and the gold standard abnormal region segmentation mask corresponding to each training sample set, the initial neural network is trained to determine the neural network.

[0123] In this step, after obtaining the predicted LGE image and the predicted anomalous region segmentation mask corresponding to each training sample set, the initial neural network is trained by combining the gold standard LGE image and the gold standard anomalous region segmentation mask corresponding to each training sample set.

[0124] In specific training, as an optional embodiment, a first loss can be calculated between the predicted LGE image of each training sample set and the corresponding gold standard LGE image; a second loss can be calculated between the predicted abnormal region segmentation mask of each training sample set and the corresponding gold standard abnormal region segmentation mask; and the initial neural network can be trained based on the first loss and the second loss to determine the neural network.

[0125] The loss functions chosen for calculating the first and second losses can be set according to the actual situation. The two loss functions can be the same or different; no specific restrictions are imposed here. The first and second losses for each training sample set can be calculated, and then the first and second losses for each training sample set can be weighted and summed or averaged to obtain the total loss. This total loss is then used to train the initial neural network, ultimately resulting in a trained neural network.

[0126] In this embodiment, the initial neural network is trained using predicted LGE images and predicted anomaly region segmentation masks from each training sample set, along with the corresponding gold-standard LGE images and gold-standard anomaly region segmentation masks. This combination of multiple different output data for training the initial neural network makes the trained network more accurate and capable of performing different tasks. Furthermore, by using multiple types of information to train the neural network, these different pieces of information can complement each other, further improving the accuracy of the trained neural network. Additionally, training the initial neural network by calculating different losses is a simple and intuitive process, thus improving the training accuracy and efficiency of the neural network.

[0127] The following detailed embodiment illustrates the technical solution of this application. (See attached image.) Figure 7 The flowchart shown above, based on the above embodiments, indicates that the method may include the following steps:

[0128] S1, acquire multiple T1-weighted images and CINE images of the tissue to be tested; these multiple T1-weighted images and CINE images were acquired without the injection of gadolinium contrast agent into the tissue to be tested;

[0129] S2, determine the corresponding T1 mapping image based on multiple T1 weighted images;

[0130] S3, perform feature extraction processing on multiple T1 weighted images to determine the first sub-feature map corresponding to the tissue to be tested;

[0131] S4, perform feature extraction processing on the T1 mapping image to determine the second sub-feature map corresponding to the tissue to be tested;

[0132] S5, perform fusion processing on the first sub-feature map and the second sub-feature map to determine the first feature map;

[0133] S6, perform fusion processing on multiple T1 weighted images, T1 mapping images, CINE images and the first feature map to determine the first fusion feature;

[0134] S7, the first fusion feature is input into the neural network through path 1 for image generation processing to generate a virtual LGE image;

[0135] S8, perform feature extraction processing on the CINE image to determine the second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during the motion process;

[0136] S9, perform fusion processing on the first feature map and the second feature map to determine the second fused feature;

[0137] S10, the second fusion feature is input into the neural network through path 2 for image segmentation processing to determine the abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes the abnormal region in the tissue under test.

[0138] In this embodiment, taking myocardium as the tissue to be tested, a neural network is mainly used to transform the quantitative but inconspicuous myocardial injury signal in the T1-weighted image and / or T1-mapping image into a virtual enhanced signal, making the myocardial injury signal visible. Simultaneously, a neural network combining image generation (i.e., a network generating a virtual LGE image) and image segmentation (i.e., a network obtaining anomaly region segmentation masks) is used to jointly predict regions in the image that may be enhanced, and to virtually enhance the image, obtaining a virtual LGE image and anomaly region segmentation masks, including an LGE high-signal myocardial segmentation mask.

[0139] As can be seen, in this embodiment, T1-weighted images are used, which include quantitative signals of damaged myocardium, providing quantitative theoretical support for the generated virtual LGE images. On the other hand, CINE images can reflect the laws of myocardial motion in the time dimension, making the generated virtual LGE images conform to the laws of actual myocardial motion. In other words, the generated virtual LGE images can ultimately be made close to the LGE images acquired when gadolinium contrast agents are actually injected.

[0140] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0141] Based on the same inventive concept, this application also provides an image generation apparatus for implementing the image generation method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more image generation apparatus embodiments provided below can be found in the limitations of the image generation method described above, and will not be repeated here.

[0142] In one embodiment, such as Figure 8 As shown, an image generation device is provided, comprising: an image acquisition module and an image generation module, wherein:

[0143] The image acquisition module is used to acquire multiple T1-weighted images and CINE images of the tissue under test;

[0144] The image generation module is used to generate a virtual LGE image corresponding to the tissue under test by performing image generation processing based on multiple T1-weighted images, CINE images, and a preset neural network. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes the gold standard LGE image.

[0145] In another embodiment, another image generation apparatus is provided. Based on the above embodiments, the image generation module may include:

[0146] An image determination unit is used to determine the corresponding T1 mapping image based on multiple T1 weighted images;

[0147] The first fusion unit is used to determine the first fusion feature corresponding to the tissue under test based on multiple T1 weighted images, T1 mapping images and CINE images;

[0148] Image generation processing is used to input the first fused features into a preset neural network for image generation processing to generate a virtual LGE image.

[0149] In another embodiment, another image generation apparatus is provided. Based on the above embodiments, the first fusion unit may include:

[0150] The feature extraction subunit is used to perform feature extraction processing on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference relationship between abnormal regions and other regions in the tissue under test.

[0151] The fusion subunit is used to fuse multiple T1-weighted images, T1 mapping images, CINE images, and the first feature map to determine the first fusion feature.

[0152] Optionally, the aforementioned feature extraction subunit is specifically used to perform feature extraction processing on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue to be tested; to perform feature extraction processing on the T1 mapping image to determine the second sub-feature map corresponding to the tissue to be tested; and to perform fusion processing on the first sub-feature map and the second sub-feature map to determine the first feature map.

[0153] In another embodiment, another image generation device is provided. Based on the above embodiment, the image generation module is specifically used to perform image generation processing based on multiple T1-weighted images, CINE images, and a preset neural network to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0154] Optionally, the image generation module described above may include:

[0155] The first feature extraction unit is used to determine the first feature map corresponding to the tissue to be tested based on multiple T1-weighted images;

[0156] The second feature extraction unit is used to perform feature extraction processing on the CINE image to determine the second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during the motion process.

[0157] The second fusion unit is used to fuse the first feature map and the second feature map to determine the second fused feature;

[0158] The segmentation unit is used to input the second fused feature into the neural network for image segmentation processing to determine the abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes the abnormal region in the tissue under test.

[0159] In another embodiment, another image generation apparatus is provided. Based on the above embodiments, the gold standard further includes a gold standard anomaly region segmentation mask. The apparatus may also include a training module, which may include:

[0160] The prediction unit is used to determine the predicted LGE image and the predicted abnormal region segmentation mask for each training image subset based on each training sample set and the initial neural network.

[0161] The training unit is used to train the initial neural network based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, and to determine the neural network.

[0162] Optionally, the above training unit may include:

[0163] The first loss calculation subunit is used to calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each training sample set.

[0164] The second loss calculation subunit is used to calculate the second loss between the predicted abnormal region segmentation mask and the corresponding gold standard abnormal region segmentation mask for each training sample set.

[0165] The training subunit is used to train the initial neural network based on the first loss and the second loss to determine the neural network.

[0166] Each module in the aforementioned image generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0167] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0168] Multiple T1-weighted images and CINE images of the tissue under test are acquired. Based on the multiple T1-weighted images, CINE images, and a pre-set neural network, image generation processing is performed to generate a virtual LGE image corresponding to the tissue under test. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes the gold standard LGE image.

[0169] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0170] The corresponding T1 mapping image is determined based on multiple T1 weighted images; the first fusion feature corresponding to the tissue under test is determined based on multiple T1 weighted images, T1 mapping images and CINE images; the first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image.

[0171] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0172] Feature extraction processing is performed on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference relationship between abnormal areas and other areas in the tissue under test; multiple T1-weighted images, T1 mapping images, CINE images and the first feature map are fused to determine the first fused feature.

[0173] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0174] Feature extraction is performed on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue under test; feature extraction is performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test; the first sub-feature map and the second sub-feature map are fused to determine the first feature map.

[0175] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0176] Image generation processing is performed using multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0177] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0178] Based on multiple T1-weighted images, a first feature map corresponding to the tissue under test is determined; feature extraction processing is performed on the CINE image to determine a second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during motion; the first feature map and the second feature map are fused to determine a second fused feature; the second fused feature is input into a neural network for image segmentation processing to determine an abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes abnormal regions in the tissue under test.

[0179] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0180] Based on each training sample set and the initial neural network, the predicted LGE image and the predicted anomaly region segmentation mask corresponding to each training image subset are determined; based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, the initial neural network is trained to determine the neural network.

[0181] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0182] Calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each training sample set; calculate the second loss between the predicted anomaly region segmentation mask and the corresponding gold standard anomaly region segmentation mask for each training sample set; train the initial neural network based on the first and second losses to determine the neural network.

[0183] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0184] Multiple T1-weighted images and CINE images of the tissue under test are acquired. Based on the multiple T1-weighted images, CINE images, and a pre-set neural network, image generation processing is performed to generate a virtual LGE image corresponding to the tissue under test. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes the gold standard LGE image.

[0185] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0186] The corresponding T1 mapping image is determined based on multiple T1 weighted images; the first fusion feature corresponding to the tissue under test is determined based on multiple T1 weighted images, T1 mapping images and CINE images; the first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image.

[0187] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0188] Feature extraction processing is performed on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference relationship between abnormal areas and other areas in the tissue under test; multiple T1-weighted images, T1 mapping images, CINE images and the first feature map are fused to determine the first fused feature.

[0189] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0190] Feature extraction is performed on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue under test; feature extraction is performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test; the first sub-feature map and the second sub-feature map are fused to determine the first feature map.

[0191] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0192] Image generation processing is performed using multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0193] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0194] Based on multiple T1-weighted images, a first feature map corresponding to the tissue under test is determined; feature extraction processing is performed on the CINE image to determine a second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during motion; the first feature map and the second feature map are fused to determine a second fused feature; the second fused feature is input into a neural network for image segmentation processing to determine an abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes abnormal regions in the tissue under test.

[0195] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0196] Based on each training sample set and the initial neural network, the predicted LGE image and the predicted anomaly region segmentation mask corresponding to each training image subset are determined; based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, the initial neural network is trained to determine the neural network.

[0197] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0198] Calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each training sample set; calculate the second loss between the predicted anomaly region segmentation mask and the corresponding gold standard anomaly region segmentation mask for each training sample set; train the initial neural network based on the first and second losses to determine the neural network.

[0199] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0200] Multiple T1-weighted images and CINE images of the tissue under test are acquired. Based on the multiple T1-weighted images, CINE images, and a pre-set neural network, image generation processing is performed to generate a virtual LGE image corresponding to the tissue under test. In the virtual LGE image, the signal contrast of abnormal areas in the tissue under test is higher than that of other areas. The neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1-weighted images, training T1 mapping images, and training CINE images. The gold standard includes the gold standard LGE image.

[0201] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0202] The corresponding T1 mapping image is determined based on multiple T1 weighted images; the first fusion feature corresponding to the tissue under test is determined based on multiple T1 weighted images, T1 mapping images and CINE images; the first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image.

[0203] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0204] Feature extraction processing is performed on multiple T1-weighted images and / or T1 mapping images to determine the first feature map corresponding to the tissue under test; the first feature map is used to characterize the difference relationship between abnormal areas and other areas in the tissue under test; multiple T1-weighted images, T1 mapping images, CINE images and the first feature map are fused to determine the first fused feature.

[0205] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0206] Feature extraction is performed on multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue under test; feature extraction is performed on the T1 mapping image to determine the second sub-feature map corresponding to the tissue under test; the first sub-feature map and the second sub-feature map are fused to determine the first feature map.

[0207] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0208] Image generation processing is performed using multiple T1-weighted images, CINE images, and a pre-set neural network to generate a virtual LGE image and an abnormal region segmentation mask corresponding to the tissue under test.

[0209] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0210] Based on multiple T1-weighted images, a first feature map corresponding to the tissue under test is determined; feature extraction processing is performed on the CINE image to determine a second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during motion; the first feature map and the second feature map are fused to determine a second fused feature; the second fused feature is input into a neural network for image segmentation processing to determine an abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes abnormal regions in the tissue under test.

[0211] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0212] Based on each training sample set and the initial neural network, the predicted LGE image and the predicted anomaly region segmentation mask corresponding to each training image subset are determined; based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, the initial neural network is trained to determine the neural network.

[0213] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0214] Calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each training sample set; calculate the second loss between the predicted anomaly region segmentation mask and the corresponding gold standard anomaly region segmentation mask for each training sample set; train the initial neural network based on the first and second losses to determine the neural network.

[0215] It should be noted that all data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are data that have been fully authorized by all parties.

[0216] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0217] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0218] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An image generation method, characterized in that, The method includes: Multiple T1-weighted images and CINE images of the tissue under test are acquired; the multiple T1-weighted images include T1-weighted images obtained after calculation of different echo data; The corresponding T1 mapping image is determined based on the multiple T1 weighted images; Feature extraction processing is performed on the multiple T1-weighted images to determine the first sub-feature map corresponding to the tissue to be tested; The T1 mapping image is subjected to feature extraction processing to determine the second sub-feature map corresponding to the tissue to be tested; The first sub-feature map and the second sub-feature map are fused to determine the first feature map; the first feature map is used to characterize the difference between abnormal regions and other regions in the tissue under test. The multiple T1-weighted images, the T1 mapping image, the CINE image, and the first feature map are fused to determine the first fused feature corresponding to the tissue under test. The first fusion feature is input into a preset neural network for image generation processing to generate a virtual LGE image corresponding to the tissue under test. In the virtual LGE image, the signal contrast of the abnormal region in the tissue under test is higher than that of other regions; the neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images for the same tissue sample. The gold standard includes the gold standard LGE image.

2. The method according to claim 1, characterized in that, The step of generating a virtual LGE image corresponding to the tissue under test by performing image generation processing based on the multiple T1-weighted images, the CINE image, and a preset neural network further includes: Image generation processing is performed based on the multiple T1-weighted images, the CINE image, and a preset neural network to generate the virtual LGE image and the abnormal region segmentation mask corresponding to the tissue under test.

3. The method according to claim 2, characterized in that, The step of generating an image generation mask for the abnormal region corresponding to the tissue under test by performing image generation processing based on the multiple T1-weighted images, the CINE image, and a preset neural network includes: Based on the multiple T1-weighted images, determine the first feature map corresponding to the tissue to be tested; The CINE image is subjected to feature extraction processing to determine the second feature map corresponding to the tissue under test; the second feature map is used to characterize the abnormal information and contour information of the tissue under test during the motion process; The first feature map and the second feature map are fused to determine the second fused feature; The second fusion feature is input into the neural network for image segmentation processing to determine the abnormal region segmentation mask corresponding to the tissue under test; the abnormal region segmentation mask includes the abnormal region in the tissue under test.

4. The method according to claim 1, characterized in that, The gold standard also includes a gold standard anomaly region segmentation mask, and the training methods for the neural network include: Based on each training sample set and the initial neural network, determine the predicted LGE image and the predicted abnormal region segmentation mask corresponding to each training sample set; The initial neural network is trained based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, thereby determining the neural network.

5. The method according to claim 4, characterized in that, The step of training the initial neural network based on the predicted LGE image, the gold standard LGE image, the predicted anomaly region segmentation mask, and the gold standard anomaly region segmentation mask corresponding to each training sample set, and determining the neural network, includes: Calculate the first loss between the predicted LGE image and the corresponding gold standard LGE image for each of the training sample sets; Calculate the second loss between the predicted anomaly region segmentation mask and the corresponding gold standard anomaly region segmentation mask for each training sample set; The initial neural network is trained based on the first loss and the second loss to determine the neural network.

6. An image generation apparatus, characterized in that, The device includes: The image acquisition module is used to acquire multiple T1-weighted images and CINE images of the tissue under test; the multiple T1-weighted images include T1-weighted images obtained after calculation of different echo data; An image generation module is used to perform feature extraction processing on the multiple T1-weighted images to determine a first sub-feature map corresponding to the tissue under test; perform feature extraction processing on the T1 mapping image to determine a second sub-feature map corresponding to the tissue under test; perform fusion processing on the first sub-feature map and the second sub-feature map to determine a first feature map; the first feature map is used to characterize the difference relationship between abnormal regions and other regions in the tissue under test; perform fusion processing on the multiple T1-weighted images, the T1 mapping image, the CINE image, and the first feature map to determine a first fusion feature corresponding to the tissue under test; and input the first fusion feature into a preset neural network for image generation processing to generate a virtual LGE image corresponding to the tissue under test. In the virtual LGE image, the signal contrast of the abnormal region in the tissue under test is higher than that of other regions; the neural network is trained based on multiple training sample sets and the gold standard corresponding to each training sample set. Each training sample set includes multiple training T1 weighted images, training T1 mapping images and training CINE images for the same tissue sample. The gold standard includes the gold standard LGE image.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images