A method, system, terminal, and medium for enhancing cardiac magnetic resonance imaging (MRI) images.

By constructing a model for generating enhanced cardiac magnetic resonance images, enhanced images of a specified contrast category can be generated based on plain cardiac magnetic resonance cinematic images. This solves the problem of not being able to obtain enhanced images in existing technologies and enables the assistance of myocardial infarction diagnosis and screening without the use of contrast agents.

CN117671063BActive Publication Date: 2026-11-17SHANGHAI TECH UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311676518.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-07
Publication Date
2026-11-17
Estimated Expiration
2043-12-07

Smart Images

  • Figure CN117671063B_ABST
    Figure CN117671063B_ABST
Patent Text Reader

Abstract

This application provides a method, system, terminal, and medium for generating enhanced cardiac magnetic resonance images. By constructing an enhanced cardiac magnetic resonance image generation model, and based on the model, generating enhanced cardiac magnetic resonance images with a specified contrast category from acquired plain scan cardiac magnetic resonance cinematic images. The enhanced cardiac magnetic resonance images generated by this invention exhibit high consistency with real gadolinium-enhanced delayed-enhance cardiac magnetic resonance images in terms of contrast, scar size, and scar transmural properties between infarcted and healthy myocardium. Therefore, it can distinguish between healthy and infarcted myocardium without contrast agent injection, assisting doctors in diagnosing myocardial infarction and conducting effective myocardial infarction screening. This not only avoids the safety risks associated with contrast agent use in patients with renal insufficiency or those at risk of allergies, but also assists in the initial screening and localization of infarcted myocardium in some emergency situations, facilitating further diagnosis and treatment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and in particular to a method, system, terminal, and medium for generating enhanced cardiac magnetic resonance images. Background Technology

[0002] Acute myocardial infarction (AM) is a common coronary artery disease caused by myocardial ischemia due to coronary artery blockage. Rapid diagnosis of the infarct location and extent is crucial for optimizing treatment plans in the diagnosis and treatment of AM. Cardiac magnetic resonance imaging (MRI) is a non-invasive imaging technique that provides high-resolution images of the heart while simultaneously assessing cardiac function, myocardial blood supply, and myocardial damage. In the diagnosis and treatment of AM, cardiac MRI can detect pathological changes such as myocardial edema, myocardial fibrosis, and myocardial necrosis, thus helping physicians determine the location, extent, and severity of the infarction.

[0003] Gadolinium-enhanced delayed enhancement imaging (GAE) is a commonly used technique in cardiac magnetic resonance imaging (MRI). Because GAE enhances the signal intensity in the infarcted area, it can detect the location, extent, and severity of myocardial infarction in patients with acute myocardial infarction, helping clinicians develop treatment plans. The diagnostic criteria for GAE primarily include the degree of gadolinium contrast agent absorption, the signal intensity of the infarcted area, and the area of ​​delayed enhancement.

[0004] Gadolinium-contrast delayed-enhancement cardiac magnetic resonance imaging (MRI) has become the gold standard for diagnosing myocardial infarction due to its high accuracy in visualizing infarct areas. However, gadolinium contrast agents cannot be used in patients with renal insufficiency and may cause adverse reactions in patients with a history of allergies. Therefore, obtaining effective descriptive information on both healthy and infarcted myocardium while avoiding gadolinium injection is of significant clinical importance.

[0005] Currently, common techniques for obtaining enhanced cardiac MRI images from contrast-free plain cardiac MRI cine images mainly include using neural networks to extract texture and motion features from the images to predict the location and size of infarct scars, and using quantitative MRI to analyze the infarct area. However, training these neural networks often requires manual annotation of the infarct area, which is time-consuming and labor-intensive. Furthermore, it cannot provide evidence of the degree and extent of myocardial infarction, making it unsuitable for clinical application. Combining plain cardiac MRI cine images and quantitative parametric images to generate images similar to gadolinium-enhanced delayed enhancement (GEM) images shows good consistency with GEM images in assessing scar size and transmural properties. Moreover, the training process does not require manual annotation of the infarct area, significantly reducing the difficulty of constructing the training dataset. The interpretation of the generated enhanced images is similar to that of GEM images, facilitating clinical analysis and diagnosis, and providing a feasible alternative to GEM images. However, this technique still requires quantitative parametric images as an auxiliary modality, and obtaining these images is complex and not a commonly used technique for examining patients with acute myocardial infarction. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method, system, terminal and medium for generating enhanced cardiac magnetic resonance images, so as to solve the problem that enhanced cardiac magnetic resonance images cannot be obtained based solely on plain scan cardiac magnetic resonance cine images in the prior art.

[0007] To achieve the above and other related objectives, a first aspect of this application provides a method for generating enhanced cardiac magnetic resonance images, the method comprising:

[0008] Constructing a model for generating enhanced cardiac magnetic resonance images;

[0009] Based on the enhanced cardiac magnetic resonance image generation model, an enhanced cardiac magnetic resonance image with a specified contrast category is generated from the acquired plain scan cardiac magnetic resonance cinematic image.

[0010] The construction method of the enhanced cardiac magnetic resonance image generation model includes:

[0011] Multiple patient image samples were acquired; each patient image sample included: plain scan cardiac magnetic resonance cine image and gadolinium-enhanced delayed-enhancement cardiac magnetic resonance image;

[0012] Layer matching was performed on each scan layer of the plain cardiac magnetic resonance cine image and the gadolinium-delayed enhancement cardiac magnetic resonance image in each patient image sample to ensure that the paired scan layer images have a consistent cardiac cross-sectional position.

[0013] For each matched patient image sample, plain cardiac magnetic resonance cine images and gadolinium-enhanced cardiac magnetic resonance images were subjected to image preprocessing based on the left ventricle as the center, with fixed resolution and pixel values ​​within a set range.

[0014] Based on the contrast features of the left ventricular blood pool region extracted from the gadolinium-delayed enhancement cardiac magnetic resonance images in each patient image sample, contrast classification is performed on each patient image sample after image preprocessing to construct a patient image sample training set.

[0015] The enhanced cardiac magnetic resonance image generation model is trained based on the patient image sample training set.

[0016] In some embodiments of the first aspect of this application, the layer matching of each scan layer of the plain cardiac magnetic resonance cine image and the gadolinium-delayed enhancement cardiac magnetic resonance image in each patient image sample includes:

[0017] Image samples from each patient were screened to remove plain scan cardiac magnetic resonance cine images with significant differences in imaging axis direction, as well as corresponding gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images.

[0018] The projection values ​​of the origin of each scan layer of the plain cardiac magnetic resonance cine image in each patient image sample after screening to the imaging axis of the corresponding gadolinium-enhanced cardiac magnetic resonance image are calculated to obtain the two scan layer images with the closest position in the plain cardiac magnetic resonance cine image and the corresponding gadolinium-enhanced cardiac magnetic resonance image.

[0019] The affine transformation relationship between the two closest scan layers in the obtained plain scan cardiac magnetic resonance cinema image and the corresponding gadolinium-enhanced cardiac magnetic resonance image is calculated respectively, so as to enable rigid registration of the two scan images.

[0020] Perform non-rigid registration on two rigidly registered scanned images.

[0021] In some embodiments of the first aspect of this application, the image preprocessing performed on the plain cardiac magnetic resonance cine images and gadolinium-enhanced cardiac magnetic resonance images in each matched patient image sample, based on a left ventricle-centered image with fixed resolution and pixel values ​​within a set range, includes:

[0022] For each matched patient image sample, the plain cardiac magnetic resonance cinematic image and the corresponding gadolinium-enhanced delayed cardiac magnetic resonance image were resampled at a fixed resolution for each scan layer.

[0023] Based on a pre-trained cardiac localization network, the resampled images of each scan layer are automatically localized and cropped to preserve the region of interest centered on the left ventricle.

[0024] The cropped scan layer images are normalized to ensure that all pixel values ​​fall within a set range.

[0025] In some embodiments of the first aspect of this application, the method of training the enhanced cardiac magnetic resonance image generation model based on the patient image sample training set includes:

[0026] A deep adversarial network is trained based on a patient image sample training set to obtain a primary enhanced cardiac magnetic resonance image generation model, and a corresponding enhanced cardiac magnetic resonance image is generated based on the plain scan cardiac magnetic resonance cinema image of each sample in the patient image sample training set.

[0027] The primary enhanced cardiac magnetic resonance imaging (MRI) image generation model is optimized based on the enhanced cardiac MRI images of each sample to obtain the final enhanced cardiac MRI image generation model.

[0028] In some embodiments of the first aspect of this application, generating a corresponding enhanced cardiac magnetic resonance image based on the plain scan cardiac magnetic resonance cinematic images of each sample in the patient image sample training set includes:

[0029] Based on the dynamic, texture, and semantic feature information of the target scan layer and the images of the adjacent upper and lower scan layers of the plain scan cardiac magnetic resonance cinema images in the patient image sample training set, a feature map containing spatiotemporal information is constructed.

[0030] The feature map is adjusted in a specific direction by specifying a contrast category;

[0031] Based on the directional adjustment feature map, by matching the triggered delayed imaging time, a frame image from the target scan layer of the plain scan cardiac magnetic resonance imaging (MRI) cine image is selected as the migration source image; wherein the cardiac phase state of the image frame is most similar to the cardiac phase state of the corresponding gadolinium-delayed enhanced cardiac MRI image.

[0032] An image migration method is used to migrate the selected image frames to generate corresponding enhanced cardiac magnetic resonance images with a specified contrast category.

[0033] In some embodiments of the first aspect of this application, the method of optimizing the primary enhanced cardiac magnetic resonance image generation model based on the enhanced cardiac magnetic resonance images of each sample includes:

[0034] Contrast augmentation was performed on the gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images of each sample in the patient image sample training set to generate augmented gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images with different contrasts.

[0035] The enhanced gadolinium-enhanced cardiac magnetic resonance image corresponding to each sample, the original gadolinium-enhanced cardiac magnetic resonance image, and the generated enhanced cardiac magnetic resonance image are compared. An iterative optimization algorithm is used to optimize the primary enhanced cardiac magnetic resonance image generation model to obtain the final enhanced cardiac magnetic resonance image generation model.

[0036] In some embodiments of the first aspect of this application, generating an enhanced cardiac magnetic resonance image with a specified contrast category based on the enhanced cardiac magnetic resonance image generation model from an acquired plain scan cardiac magnetic resonance cinematic image includes:

[0037] The acquired plain scan cardiac magnetic resonance cine images were preprocessed based on the left ventricle as the center, with a fixed resolution and pixel values ​​within a set range.

[0038] By specifying the contrast category and the ECG gating trigger time, a corresponding enhanced cardiac magnetic resonance image is generated based on the pre-processed plain scan cardiac magnetic resonance cinematic image.

[0039] To achieve the above and other related objectives, a second aspect of this application provides an enhanced cardiac magnetic resonance image generation system, the system comprising:

[0040] The enhanced cardiac magnetic resonance image generation model construction module is used to perform scan layer matching, image preprocessing and contrast classification on plain scan cardiac magnetic resonance cinema images and corresponding gadolinium-delayed enhanced cardiac magnetic resonance images from multiple patient image samples to construct a patient image sample training set, and to train the enhanced cardiac magnetic resonance image generation model based on the patient image sample training set.

[0041] An enhanced cardiac magnetic resonance image generation model prediction module, connected to the enhanced cardiac magnetic resonance image generation model training module, is used to generate corresponding enhanced cardiac magnetic resonance images from preprocessed plain scan cardiac magnetic resonance cinema images based on the constructed enhanced cardiac magnetic resonance image generation model by specifying the contrast category and ECG gating trigger time.

[0042] To achieve the above and other related objectives, a third aspect of this application provides an electronic terminal, comprising: a processor and a memory; the memory for storing a computer program, and the processor for executing the computer program stored in the memory to cause the terminal to perform any of the enhanced cardiac magnetic resonance image generation methods.

[0043] To achieve the above and other related objectives, a fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods for generating enhanced cardiac magnetic resonance images.

[0044] As described above, this application provides a method, system, terminal, and medium for generating enhanced cardiac magnetic resonance images. It constructs an enhanced cardiac magnetic resonance image generation model and, based on the model, generates enhanced cardiac magnetic resonance images with a specified contrast category from acquired plain scan cardiac magnetic resonance cinematic images. The enhanced cardiac magnetic resonance images generated by this invention exhibit high consistency with real gadolinium-enhanced delayed-enhance cardiac magnetic resonance images in terms of contrast, scar size, and scar transmural properties between infarcted and healthy myocardium. Therefore, it can distinguish between healthy and infarcted myocardium without contrast agent injection, assisting doctors in diagnosing myocardial infarction and conducting effective myocardial infarction screening. This not only avoids the safety risks associated with contrast agent use in patients with renal insufficiency or those at risk of allergies, but also assists in the initial screening and localization of infarcted myocardium in some emergency situations, facilitating further diagnosis and treatment. Attached Figure Description

[0045] Figure 1 The diagram shown is a flowchart illustrating a method for generating enhanced cardiac magnetic resonance images according to an embodiment of this application.

[0046] Figure 2 The diagram shown is a flowchart illustrating a method for generating enhanced cardiac magnetic resonance images according to a specific embodiment of this application.

[0047] Figure 3 The diagram shown is a schematic of a cardiac magnetic resonance image generation model trained with a deep adversarial network in a specific embodiment of this application.

[0048] Figure 4 The diagram shown is a schematic representation of an enhanced cardiac magnetic resonance image generation system according to an embodiment of this application.

[0049] Figure 5 The diagram shown is a structural schematic of an enhanced cardiac magnetic resonance image generation terminal according to an embodiment of this application. Detailed Implementation

[0050] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0051] It should be noted that in the following description, reference is made to the accompanying drawings, which illustrate several embodiments of this application. It should be understood that other embodiments may also be used, and changes in mechanical composition, structure, electrical system, and operation may be made without departing from the spirit and scope of this application. The following detailed description should not be considered limiting, and the scope of the embodiments of this application is defined only by the claims of the published patent. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. Spatially related terms, such as “upper,” “lower,” “left,” “right,” “below,” “below,” “lower part,” “above,” “upper part,” etc., may be used herein to illustrate the relationship between one element or feature shown in the figures and another element or feature.

[0052] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," "fixing," and "holding" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0053] Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” indicate the presence of the stated feature, operation, element, component, item, kind, and / or group, but do not preclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” Exceptions to this definition arise only when combinations of elements, functions, or operations are inherently mutually exclusive in some manner.

[0054] To address the problems described in the background section, this invention provides a method, system, terminal, and medium for generating enhanced cardiac magnetic resonance images, aiming to solve the problem in the prior art that enhanced cardiac magnetic resonance images cannot be obtained solely from plain cardiac magnetic resonance cine images. Meanwhile, to make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention are further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0055] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0056] <1> DICOM (Digital Imaging and Communications in Medicine) is an international standard (ISO 12052) for medical images and related information, defining a medical image format that meets clinical needs and can be used for data exchange.

[0057] <2> FOV (Field of View): Imaging field of view;

[0058] <3> SSD (Single Shot Multi Box Detector): Single-stage object detection is a detection method that uses convolutional neural networks to detect objects in images. It is easy to train and can be directly integrated into systems that require detection components.

[0059] like Figure 1 The diagram illustrates a flowchart of a method for generating enhanced cardiac magnetic resonance images according to an embodiment of the present invention. The method in this embodiment mainly includes the following steps:

[0060] Step S1: Construct a model for generating enhanced cardiac magnetic resonance images.

[0061] Step S1 mainly includes:

[0062] Step S11: Acquire multiple patient image samples; wherein each patient image sample includes: plain scan cardiac magnetic resonance cinematic image and gadolinium-enhanced delayed-enhancement cardiac magnetic resonance image.

[0063] Step S12: Perform layer matching on each scan layer of the plain scan cardiac magnetic resonance cine image and the gadolinium-delayed enhancement cardiac magnetic resonance image in each patient image sample, so that the two scan layer images after pairing have a consistent cardiac cross-sectional position.

[0064] It is understandable that plain MRI cine images of the heart are a set of dynamic images based on multiple scan layers, with each scan layer corresponding to one dynamic image. Each dynamic image continuously depicts the contraction and relaxation of the heart within a cardiac cycle, and each frame corresponds to a cardiac phase. Gadolinium-enhanced delayed-enhanced MRI images of the heart, on the other hand, are a set of static images based on multiple scan layers, with each scan layer corresponding to one static image, describing only a specific cardiac phase. The images in both plain MRI cine images and gadolinium-enhanced delayed-enhanced MRI images describe different cross-sectional positions of the heart, presenting the geometric contours, dynamics, texture, and semantic features of the heart based on multiple parallel cross-sections. Images with consistent cardiac cross-sectional positions refer to cross-sectional views of the heart described from the same or similar positions and angles, where the geometric contours, dynamics, texture, and semantic features of the heart are consistent.

[0065] In this embodiment of the invention, the correspondence between the plain scan cardiac magnetic resonance cinema images and the corresponding gadolinium-enhanced cardiac magnetic resonance images in each patient image sample is analyzed. Plain scan and enhanced scan layer images with consistent cardiac cross-sectional positions are matched to form plain scan cardiac magnetic resonance cinema image-gadolinium-enhanced cardiac magnetic resonance image data pairs, thereby improving the training speed and efficiency of the enhanced cardiac magnetic resonance image generation model.

[0066] In one embodiment, step S12 includes:

[0067] Step S121: Screen the image samples of each patient to remove plain scan cardiac magnetic resonance cine images with too large differences in imaging axis direction and the corresponding gadolinium-enhanced cardiac magnetic resonance images.

[0068] When the imaging axis of a plain cardiac magnetic resonance imaging (MRI) cine image and a gadolinium-enhanced cardiac MRI image differs significantly (i.e., the cardiac sections in the two images are not parallel and have angular differences), the scanned layers cannot be matched to a consistent cardiac section position. This invention effectively improves the efficiency and accuracy of layer matching by filtering out invalid data before performing the layer matching operation. It also saves training time and improves training efficiency and accuracy for the overall construction of the enhanced cardiac MRI image generation model.

[0069] Specifically, this invention can use cosine similarity to evaluate the angle between the imaging axis of a plain scan cardiac MRI cine image and the imaging axis of a gadolinium-enhanced cardiac MRI image. The cosine similarity is compared with a preset minimum cosine similarity threshold. If the cosine similarity is less than the preset minimum cosine similarity threshold, the corresponding plain scan cardiac MRI cine image and gadolinium-enhanced cardiac MRI image are rejected. If the cosine similarity is not less than the preset minimum cosine similarity threshold, the corresponding plain scan cardiac MRI cine image and gadolinium-enhanced cardiac MRI image are accepted and layer matching continues. It should be noted that the minimum cosine similarity threshold can be set according to requirements, and this invention does not impose any limitations on it.

[0070] Step S122: Calculate the projection value of the origin of each scan layer of the plain cardiac magnetic resonance cine image in each patient image sample after screening to the imaging axis of the corresponding gadolinium-enhanced cardiac magnetic resonance image, so as to obtain the two scan layer images with the closest position in the plain cardiac magnetic resonance cine image and the corresponding gadolinium-enhanced cardiac magnetic resonance image.

[0071] Specifically, this invention extracts metadata from the DICOM files of plain scan cardiac MRI cine images and corresponding gadolinium-enhanced cardiac MRI images to obtain the coordinate origin of each scan layer of the plain scan cardiac MRI cine image and the imaging axis direction parameters of the corresponding gadolinium-enhanced cardiac MRI image. Based on these parameters, the projection value from the coordinate origin of each scan layer of the plain scan cardiac MRI cine image to the imaging axis of the corresponding gadolinium-enhanced cardiac MRI image is calculated. The projection value of each plain scan layer image is compared with the limiting layer pairing threshold. If the projection value of a certain plain scan layer image is greater than the limiting layer pairing threshold, then the plain scan layer image cannot be matched with any gadolinium-enhanced cardiac MRI image within the threshold distance, and the plain scan layer image is rejected. Otherwise, based on the projection value of the plain scan layer image, the enhancement scan layer image with the closest position in the corresponding gadolinium-enhanced cardiac MRI image is matched, thereby completing the layer matching of the plain scan cardiac MRI cine image and the corresponding gadolinium-enhanced cardiac MRI image. It should be noted that the limit layer pairing threshold can be set according to requirements, and the present invention does not impose any limitations on it.

[0072] Step S123: Calculate the affine transformation relationship between the two closest scan layers in the obtained plain cardiac magnetic resonance cine image and the corresponding gadolinium-enhanced cardiac magnetic resonance image, respectively, for rigid registration of the two scan images; and non-rigid registration.

[0073] First, this invention calculates the affine transformation relationship between two paired plain scan layer images and enhanced scan layer images by comparing their coordinate system parameters; wherein the affine transformation relationship includes affine parameters and inverse operation parameters. Specifically, this invention extracts the metadata of the DICOM files of the paired plain scan layer images and enhanced scan layer images to obtain the pixel size and field of view (FOV) parameters of the two scan layer images, and calculates the affine transformation relationship between the images by comparing their coordinate system parameters based on these parameters.

[0074] Secondly, based on the calculation, the present invention obtains the affine transformation relationship of two paired scan layer images and performs rigid registration on the two scan layer images.

[0075] Step S124: Perform non-rigid registration on the two scanned images that have been rigidly registered.

[0076] The purpose of this design in this embodiment of the invention is to eliminate the differences in FOV parameters and voxel size between the plain scan image and the enhanced scan image by performing rigid and non-rigid registration on the matched plain scan image and the gadolinium-delayed enhanced scan image. It can also overcome the non-rigid image deformation caused by cardiac contraction and relaxation and slight patient movement during the imaging process, thereby achieving accurate registration and making the plain scan image and the corresponding enhanced scan image consistent in imaging position, that is, having a consistent cardiac cross-sectional position.

[0077] In one embodiment, step S12 further includes: rejecting plain scan layer images and enhanced scan layer images that failed to match.

[0078] Many failed image pairings are usually caused by patient movement during the scan. The scan coordinates recorded in their DICOM file information become inaccurate due to patient movement, resulting in geometrically misaligned pairing data. Therefore, by checking the DICOM file information, failed image pairings can be rejected to avoid interfering with subsequent model training.

[0079] Step S13: Perform image preprocessing on the plain scan cardiac magnetic resonance cinematic image and the gadolinium-enhanced delayed-enhancement cardiac magnetic resonance image in each matched patient image sample, based on the left ventricle as the center, with a fixed resolution and pixel values ​​within a set range.

[0080] The purpose of this design in this embodiment of the invention is to ensure that the training data of the constructed enhanced cardiac magnetic resonance image generation model maintains a uniform physical size and FOV parameter, so that the enhanced cardiac magnetic resonance image generation model constructed based on the image data after image preprocessing can achieve better training and prediction results.

[0081] In one embodiment, step S13 includes:

[0082] Step S131: Resample each scan layer of the plain cardiac magnetic resonance cinematic image and the corresponding gadolinium-delayed enhancement cardiac magnetic resonance image in each matched patient image sample based on a fixed resolution.

[0083] Specifically, the present invention can resample paired flat scan layer images and enhanced scan layer images according to the ratio between the physical resolution of the original image and the physical resolution of the target image, so as to unify the physical resolution of all scan layer images.

[0084] Step S132: Based on the pre-trained cardiac localization network, automatically locate and crop the resampled images of each scan layer to retain the region of interest centered on the left ventricle.

[0085] This invention utilizes a pre-trained cardiac localization network to predict the heart position in each scan layer of all resampled plain MRI cine images and gadolinium-enhanced delayed-enhancement MRI images, obtaining a localization box describing the location of the left ventricle. The scan layer images are then cropped around this localization box to obtain a region of interest centered on the left ventricle. The target image size for cropping can be calculated by combining preset FOV parameters and the physical resolution of the image, ensuring that the obtained region of interest has a fixed FOV parameter. It should be noted that the preset FOV parameter and image resolution can be set according to requirements, and this invention does not impose any limitations.

[0086] The cardiac localization network employs the SSD method, based on a feedforward convolutional network, to automatically locate the heart in cardiac magnetic resonance images. The network performs object detection on the input short-axis cardiac magnetic resonance images, outputting a left ventricular bounding box. The center position and size of this bounding box are continuously adjusted based on the box's annotation information to reduce the error between the predicted and ground truth bounding boxes, until the model parameters converge, completing the training of the cardiac localization network. The purpose of this design in this embodiment is to reduce the workload of manually annotating the heart's location, achieve automatic prediction, and make it more suitable for clinical application.

[0087] Step S133: Normalize the cropped images of each scan layer to ensure that all pixel values ​​fall within a set range. It should be noted that the pixel value range can be set according to requirements, and this invention does not impose any limitations on it.

[0088] Step S14: Based on the contrast features of the left ventricular blood pool region extracted from the gadolinium-delayed enhancement cardiac magnetic resonance image in each patient image sample, perform contrast classification on each patient image sample after image preprocessing to construct a patient image sample training set.

[0089] Because gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images exhibit varying brightness during imaging, they display different contrast characteristics, primarily including bright blood, gray blood, and black blood contrast. Therefore, this invention uses the brightness of the left ventricular blood pool region in the pre-processed gadolinium-enhanced delayed-enhancement image as a feature index to classify the contrast of each enhanced scan layer image and its corresponding matching plain scan layer image; that is, assigning a contrast category to each enhanced scan layer image and its corresponding matching plain scan layer image. Specifically, this invention employs a mean clustering algorithm to perform cluster analysis on the brightness values ​​of each enhanced scan layer image. By finding local points in the number of categories versus average distance graph, the optimal number of cluster centers is determined, ensuring the accuracy and efficiency of the classification.

[0090] By classifying the contrast of each patient image sample and adding contrast categories, a patient image sample training set is constructed accordingly. The purpose of this design in this embodiment is to support the subsequent contrast augmentation of the original gadolinium-enhanced cardiac magnetic resonance images when training the enhanced cardiac magnetic resonance image generation model, and to effectively improve the enhanced cardiac magnetic resonance image generation model trained based on this training data, which can generate enhanced cardiac magnetic resonance images with similar contrast to real gadolinium-enhanced cardiac magnetic resonance images. This provides more accurate enhancement information to assist doctors in diagnosing the location, degree, and extent of myocardial infarction without requiring doctors to change their original work habits.

[0091] Step S15: Train an enhanced cardiac magnetic resonance image generation model based on the patient image sample training set.

[0092] In one embodiment, step S15 includes:

[0093] Step S151: Train a deep adversarial network based on the patient image sample training set to obtain a primary enhanced cardiac magnetic resonance image generation model, and generate the corresponding enhanced cardiac magnetic resonance image based on the plain scan cardiac magnetic resonance cinema image of each sample in the patient image sample training set.

[0094] In a preferred embodiment, the enhanced cardiac MRI image generation model is trained using a deep adversarial network. The deep adversarial network comprises two networks: a generator and a discriminator, which learn simultaneously. The generator is used to generate corresponding enhanced cardiac MRI images from plain scan cardiac MRI cinematic images. The generator can be designed as a multi-scale model based on a U-Net structure; convolutional long short-term memory modules or attention mechanisms can also be considered; however, this invention does not impose specific limitations.

[0095] In one embodiment, step S151 includes:

[0096] Step S1511: Based on the dynamic, texture, and semantic feature information of the target scan layer and the images of the adjacent upper and lower scan layers of the plain scan cardiac magnetic resonance cinematic images in the patient image sample training set, construct a feature map containing spatiotemporal information.

[0097] Specifically, this invention employs a multi-level spatiotemporal encoder to extract dynamic, textural, and semantic feature information from the scan layers of plain-scan cardiac magnetic resonance imaging (MRI) cine images in the patient image sample training set, thereby constructing a feature map with spatiotemporal characteristics. The multi-level spatiotemporal encoder supports the simultaneous extraction of feature information from multiple scan layers. That is, when processing the target scan layer image, adjacent scan layer images can be used as contextual reference information to simultaneously extract the spatiotemporal features of the target scan layer image and the adjacent upper and lower scan layer images. An attention mechanism is then used to fuse the spatiotemporal features of the three scan layers, thereby enabling effective feature extraction and achieving virtual enhancement of the target scan layer image. The spatiotemporal features mainly include the temporal and spatial relationships of cardiac morphology, structure, and texture, implicitly containing information about the cardiac infarction pattern, infarction extent, and infarction degree.

[0098] Step S1512: Adjust the feature map in a specific direction by specifying the contrast category.

[0099] Specifically, this invention specifies the contrast category of the target scan layer image of a plain cardiac magnetic resonance imaging (MRI) cine image for virtual enhancement, and uses conditional normalization technology to directionally adjust the feature map to generate an enhanced cardiac MRI image that conforms to a specific contrast style. For example, by specifying the use of a contrast category representing bright blood as the generation condition, the feature map will be modulated to a specific mean and variance, thereby making the generated enhanced cardiac MRI image conform to the contrast characteristics of a bright blood gadolinium-enhanced cardiac MRI image.

[0100] This invention uses the contrast category of a gadolinium-enhanced cardiac magnetic resonance image as a condition for generating the enhanced cardiac magnetic resonance image. In other words, the contrast category is used as prior information for generating the image. This enables the generation of enhanced cardiac magnetic resonance images with contrast similar to the target gadolinium-enhanced cardiac magnetic resonance image, thus avoiding training failure due to contrast differences with the target gadolinium-enhanced cardiac magnetic resonance image and improving training efficiency.

[0101] Step S1513: Based on the feature map of the orientation adjustment, by matching the triggered delayed imaging time, select a frame image from the target scan layer image of the plain scan cardiac magnetic resonance cinema image as the migration source image; wherein the cardiac phase state of the image frame is most similar to the cardiac phase state of the corresponding gadolinium-delayed enhanced cardiac magnetic resonance image.

[0102] Plain scan cardiac MRI cine images contain various cardiac phases throughout the cardiac cycle, while gadolinium-enhanced cardiac MRI, using ECG-gated triggering, is a set of static images containing only one cardiac phase. This invention selects a cine image frame from the target scan layer of the plain scan cardiac MRI cine image that most closely matches the cardiac phase of the corresponding gadolinium-enhanced cardiac MRI image as the migration source image. This ensures that the enhanced cardiac MRI image generated based on the cine image frame has a consistent cardiac geometric contour with the target gadolinium-enhanced cardiac MRI image.

[0103] Step S1514: Using an image migration method, the selected movie image frames are migrated to generate corresponding enhanced cardiac magnetic resonance images with a specified contrast category.

[0104] Specifically, this invention employs a static image style transfer network to convert the selected movie image frames into enhanced cardiac magnetic resonance images. The static image style transfer network uses spatiotemporal features obtained from a multi-level spatiotemporal encoder as conditions. Specifically, the spatiotemporal features of images from the same scan layer are applied to each scale level of the static transfer network.

[0105] The purpose of this invention in this embodiment is to transfer a static cardiac film structure image into an enhanced image that effectively depicts the infarction pattern, thereby generating enhanced cardiac magnetic resonance images without the injection of contrast agents. This helps to distinguish between healthy myocardium and infarcted myocardium, and the generated enhanced cardiac magnetic resonance images are of high quality, exhibiting a high degree of consistency with real gadolinium-enhanced delayed enhancement images in terms of contrast, scar size, and scar transmural properties between infarcted and healthy myocardium.

[0106] Step S152: Optimize the primary enhanced cardiac magnetic resonance image generation model based on the enhanced cardiac magnetic resonance images of each sample to obtain the final enhanced cardiac magnetic resonance image generation model.

[0107] In one embodiment, step S152 includes:

[0108] Step S1521: Perform contrast augmentation on the gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images of each sample in the patient image sample training set to generate augmented gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images with different contrasts.

[0109] Specifically, this invention employs an image transfer method to augment gadolinium-enhanced delayed-enhancement (DELETE) images. Given the scarcity of DELETE cardiac MRI images from patients with acute myocardial infarction, and the limited sample size easily leading to overfitting on the training set and thus degrading model prediction performance, this embodiment uses the contrast labels of the original DELETE cardiac MRI images as category markers. An image transfer network is used to randomly generate augmented DELETE cardiac MRI images with different contrast labels. Furthermore, a data augmentation pipeline is constructed by applying random flipping, random rotation, and random affine transformations. This significantly expands the limited data sample, preventing the model from overfitting during training and improving the prediction quality of the generated enhanced cardiac MRI images.

[0110] Step S1522: Compare the augmented gadolinium-enhanced cardiac magnetic resonance image corresponding to each sample with the original gadolinium-enhanced cardiac magnetic resonance image and the generated enhanced cardiac magnetic resonance image, and use an iterative optimization algorithm to optimize the primary enhanced cardiac magnetic resonance image generation model to obtain the final enhanced cardiac magnetic resonance image generation model.

[0111] In a preferred embodiment, the deep adversarial network architecture used to train the enhanced cardiac MRI image generation model includes a discriminator. The discriminator can distinguish between enhanced cardiac MRI images generated based on the model and target gadolinium-enhanced cardiac MRI images, and provide a discrimination score. A higher discrimination score indicates that the discriminator considers the input image more likely to be a genuine gadolinium-enhanced image, while a lower score indicates that the discriminator considers the image more likely to be a generated enhanced image. The target gadolinium-enhanced cardiac MRI image includes the original gadolinium-enhanced cardiac MRI image for each sample in the patient image sample training set and the augmented gadolinium-enhanced cardiac MRI image generated after contrast augmentation.

[0112] This invention employs a loss function to evaluate the performance of the enhanced cardiac magnetic resonance imaging (MRI) image generation model and determines the quality of the generated enhanced cardiac MRI images based on the loss. The loss function is expressed as a weighted sum of adversarial loss and image similarity loss, as shown in formula (1); the adversarial loss is denoted as... Its calculation depends on the discrimination scores given by the discriminator for the generated enhanced cardiac magnetic resonance image and the target rolled delayed enhanced cardiac magnetic resonance image, as shown in formula (2); the image similarity loss is denoted as The calculation is based on the pixel value difference, semantic difference, and statistical difference between the generated enhanced cardiac magnetic resonance image and the target rolled delayed enhanced cardiac magnetic resonance image, as shown in formula (3). When the loss function no longer decreases significantly with iteration and the generated enhanced cardiac magnetic resonance image tends to stabilize, it can be determined that the enhanced cardiac magnetic resonance image generation model has converged.

[0113]

[0114]

[0115]

[0116] Where G represents the generator, D represents the discriminator, x is the generated enhanced cardiac MRI image, and y is the target gadolinium-enhanced delayed-enhance cardiac MRI image. To combat the losses, For image similarity loss, For L1 loss, In order to perceive loss, The mutual information loss is represented by λ1, λ2, and λ3, which are linear weighting parameters.

[0117] In one specific embodiment, the present invention employs two Adam optimizers to alternately optimize the discriminator and the generator until the training of the enhanced cardiac MRI images converges. Convergence is indicated when the discriminator can no longer clearly distinguish between the generated enhanced cardiac MRI images and the target gadolinium-enhanced delayed-enhance cardiac MRI images, and when the image quality generated by the generator tends to stabilize, with no further room for improvement.

[0118] Step S2: Based on the enhanced cardiac magnetic resonance image generation model, generate an enhanced cardiac magnetic resonance image with a specified contrast category from the acquired plain scan cardiac magnetic resonance cinematic image.

[0119] In one embodiment, step S2 includes:

[0120] Step S21: Perform image preprocessing on the acquired plain scan cardiac magnetic resonance cine images based on the left ventricle as the center, with a fixed resolution and pixel values ​​within a set range.

[0121] Specifically, the image preprocessing method in step S21 is the same as that in step S13. It should be noted that the range of pixel values ​​can be set according to requirements, and this invention does not impose any limitations.

[0122] Step S22: By specifying the contrast category and the ECG gating trigger time, generate the corresponding enhanced cardiac magnetic resonance image based on the plain scan cardiac magnetic resonance cinematic image after image preprocessing.

[0123] Specifically, this invention uses a depth generator to take any pre-processed plain scan cardiac magnetic resonance imaging (MRI) film image as input, and gives the contrast category and ECG gating trigger time as virtual enhancement parameters, so as to generate a generated enhanced cardiac MRI image that meets the set parameters.

[0124] To better illustrate and describe the method for generating enhanced cardiac magnetic resonance images, this invention provides a specific embodiment.

[0125] Example 1: A method for generating enhanced cardiac magnetic resonance images.

[0126] like Figure 2 as well as Figure 3 As shown, the method includes:

[0127] Step 1: Patient data collection.

[0128] The data for this study came from Renji Hospital in Shanghai. Using a Philips Ingenia 3.0T MRI scanner, data from 284 patients with myocardial infarction were acquired. Data from 221 patients were used as the training set, and data from the remaining 63 patients were used as the test set. Each patient's data included plain cine images of the heart in short-axis views acquired using a balanced steady-state free precession sequence, and gadolinium-enhanced images of the heart in short-axis views acquired using an inversion recovery sequence.

[0129] Step 2: Image scanning layer matching.

[0130] First, the cosine similarity of the angle between the imaging axis direction of each scan layer of the flat scan cinematic image and the imaging axis direction of the gadolinium-delayed enhancement image is calculated. If the cosine similarity is less than 0.95, the corresponding scan layer image is rejected.

[0131] Secondly, by analyzing DICOM metadata, relevant labels are extracted from each scan layer of the cardiac short-axis plain scan cine image and the gadolinium-enhanced delayed-enhance image. The projection value from the origin of the coordinate system of each plain scan cine image scan layer to the imaging axis of the gadolinium-enhanced delayed-enhance image is calculated and compared with a set threshold of 4 mm. If a scan layer of the plain scan cine image cannot match any gadolinium-enhanced delayed-enhance image within the threshold distance, then the necessary training data for this layer cannot be constituted, and the data for this scan layer is discarded.

[0132] Subsequently, by comparing the image coordinate system parameters between the paired plain scan cine images and gadolinium-delayed enhanced images, the affine transformation relationship from the plain scan cine image to the gadolinium-delayed enhanced image in the same plane is calculated. The plain scan cine image is rigidly registered to make it consistent with the gadolinium-delayed enhanced image in imaging position. Then, the MIND non-rigid registration algorithm is used to register the gadolinium-delayed enhanced image to the plain scan cine image that is closest to the mental state, thus achieving a precise registration process.

[0133] Finally, through manual inspection, flat scan movie images and gadolinium-delayed enhanced images that failed to pair were rejected.

[0134] Step 3: Image preprocessing.

[0135] First, with a fixed physical resolution of 0.89 × 0.89 mm... 2 All images were resampled to ensure consistent physical size for each pixel. Next, a pre-trained cardiac localization network was used to locate the left ventricle in the plain scan cinema images. Centered on the left ventricle, all images were cropped to a fixed size of 192×192 pixels, resulting in an image dataset centered on the left ventricle with a fixed field of view (FOV). Finally, all plain scan cinema images and gadolinium-enhanced images were normalized, mapping them to a range with a minimum value of 0 and a maximum value of 1.

[0136] Step 4: Image contrast classification.

[0137] The blood pool brightness in gadolinium-enhanced images was used as a feature index, and the K-means algorithm was applied to cluster the brightness values. In this embodiment, five cluster centers were selected to distinguish the contrast of gadolinium-enhanced images of different degrees. The classification results were represented by integers [1,2,3,4,5], and each delayed-enhanced image was assigned one of these five integers as a contrast label.

[0138] Step 5: Construct enhanced cardiac magnetic resonance images.

[0139] like Figure 3As shown, this embodiment constructs a generative network training framework based on deep adversarial networks, which includes three modules: contrast data augmentation, virtual augmentation generator, and adversarial network training.

[0140] In this embodiment, the contrast labels of the gadolinium-delayed enhancement images in the training set are used as category labels, and the Star GAN image transfer network is used to randomly generate augmented gadolinium-delayed enhancement images with different contrast labels.

[0141] The virtual enhancement generator receives plain scan movie images from the training set as input and performs virtual enhancement on the plain scan images of each scan layer, generating enhanced images with contrast similar to gadolinium-delayed enhancement images. This embodiment employs a five-scale spatiotemporal encoder to extract dynamic, texture, and semantic features from cardiac plain scan movie image data, forming a multi-scale image pyramid structure from large to small and from coarse to fine. At each scale, a three-dimensional convolutional module with residual connections and a convolutional long short-term memory module are used sequentially to learn the dynamic information that changes over time in the plain scan movie images, constructing feature maps containing spatiotemporal information. Finally, based on these feature maps, a static image transfer network is used to directionally generate enhanced images that conform to a specific contrast style.

[0142] In this embodiment, a discriminator is used to distinguish between generated enhanced images and gadolinium-delayed enhanced images. The discriminator employs four strided convolutional layers with a stride of 2 to achieve image sampling, resulting in a feature map of 1 / 16th the size of the original image. Specifically, at the 1 / 2 resolution scale, a self-attention layer is introduced to provide learning capabilities for global features. The final feature map size is equivalent to taking a 16×16 pixel image patch from the original input image. Further, the average feature is obtained by averaging the feature representations of all image patches through global pooling; and the average feature is scored through a linear layer and a projection layer conditioned on the category. A higher score indicates that the discriminator is more likely to consider the input image to be a genuine gadolinium-delayed enhanced image, while a lower score indicates that the discriminator is more likely to consider the image to be a generated enhanced image.

[0143] According to the training objective, this invention specifically employs two Adam optimizers to alternately optimize the discriminator and generator until training converges.

[0144] Step S6: Generate the predicted enhanced image.

[0145] In this embodiment, a depth generator is used to take any plain scan cine image from the test set as input, and scans all plain scan cine images in each scan layer using a sliding window of size 3. Simultaneously, given the contrast category and ECG gating trigger time as virtual enhancement parameters, a generated enhanced image conforming to the set parameters can be generated. By setting appropriate generation parameters, consistent cardiac state and contrast category are ensured between the generated enhanced image and the gadolinium-treated delayed enhancement image. Furthermore, the generated enhanced image and the gadolinium-treated delayed enhancement image show high consistency in contrast, scar size, and scar transmural properties between infarcted and healthy myocardium.

[0146] like Figure 4 The diagram illustrates a structural schematic of an enhanced cardiac magnetic resonance image generation system according to an embodiment of the present invention. In this embodiment, the system 400 includes:

[0147] The enhanced cardiac magnetic resonance image generation model construction module 401 is used to perform scan layer matching, image preprocessing and contrast classification on plain scan cardiac magnetic resonance cinema images and corresponding gadolinium-delayed enhanced cardiac magnetic resonance images in multiple acquired patient image samples to construct a patient image sample training set, and to train the enhanced cardiac magnetic resonance image generation model based on the patient image sample training set.

[0148] The enhanced cardiac magnetic resonance image generation model prediction module 402 is connected to the enhanced cardiac magnetic resonance image generation model training module 401. It is used to generate corresponding enhanced cardiac magnetic resonance images from the pre-processed plain scan cardiac magnetic resonance cinema images based on the constructed enhanced cardiac magnetic resonance image generation model by specifying the contrast category and the ECG gating trigger time.

[0149] It should be noted that the enhanced cardiac magnetic resonance imaging (MRI) image generation system provided in the above embodiments is only illustrated by the division of the above-described program modules. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the system can be divided into different program modules to complete all or part of the processing described above. Furthermore, the enhanced cardiac MRI image generation system and the enhanced cardiac MRI image generation method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0150] The enhanced cardiac magnetic resonance imaging (MRI) image generation method provided in this invention can be implemented on the terminal side or the server side. For details regarding the hardware structure of the enhanced cardiac MRI image generation terminal, please refer to [link to relevant documentation]. Figure 5This is a schematic diagram of an optional hardware structure of a cardiac magnetic resonance imaging (MRI) image generation terminal 500 provided in an embodiment of the present invention. The terminal 500 can be a mobile phone, computer device, tablet device, personal digital processing device, factory back-end processing device, etc. The terminal 500 includes: at least one processor 501, a memory 502, at least one network interface 504, and a user interface 506. The various components in the device are coupled together through a bus system 505. It is understood that the bus system 505 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 505 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 5 The general will label all buses as bus systems.

[0151] The user interface 506 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0152] It is understood that memory 502 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0153] In this embodiment of the invention, the memory 502 is used to store various types of data to support the operation of the terminal 500. Examples of this data include: any executable program for operation on the terminal 500, such as the operating system 5021 and application programs 5022; the operating system 5021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 5022 may contain various applications, such as a media player, browser, etc., for implementing various application services. The enhanced cardiac magnetic resonance image generation method provided in this embodiment of the invention can be included in the application program 5022.

[0154] The methods disclosed in the above embodiments of the present invention can be applied to processor 501, or implemented by processor 501. Processor 501 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 501 or by instructions in the form of software. The processor 501 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 501 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 501 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0155] In an exemplary embodiment, the terminal 500 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0156] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented using computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0157] In the embodiments provided in this application, the computer-readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, flash memory, USB flash drive, portable hard drive, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer. Additionally, any connection may be appropriately referred to as a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. However, it should be understood that computer-readable and writable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but are intended for non-transient, tangible storage media. The disks and optical discs used in the application include compact discs (CDs), laser discs, optical discs, digital multifunction discs (DVDs), floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while optical discs use lasers to copy data optically.

[0158] In summary, this application provides a method, system, terminal, and medium for generating enhanced cardiac magnetic resonance images. By constructing an enhanced cardiac magnetic resonance image generation model and based on this model, it generates enhanced cardiac magnetic resonance images with a specified contrast category from acquired plain scan cardiac magnetic resonance cinematic images. The enhanced cardiac magnetic resonance images generated by this invention exhibit high consistency with real gadolinium-enhanced delayed-enhance cardiac magnetic resonance images in terms of contrast, scar size, and scar transmural properties between infarcted and healthy myocardium. Therefore, it can distinguish between healthy and infarcted myocardium without contrast agent injection, assisting doctors in diagnosing myocardial infarction and conducting effective myocardial infarction screening. This not only avoids the safety risks associated with contrast agent use in patients with renal insufficiency or those at risk of allergies, but also assists in the initial screening and localization of infarcted myocardium in some emergency situations, facilitating further diagnosis and treatment. Therefore, this application effectively overcomes the various shortcomings of the prior art and has high industrial applicability.

[0159] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for generating enhanced cardiac magnetic resonance images, characterized in that, The method includes: Constructing a model for generating enhanced cardiac magnetic resonance images; Based on the enhanced cardiac magnetic resonance image generation model, an enhanced cardiac magnetic resonance image with a specified contrast category is generated from the acquired plain scan cardiac magnetic resonance cinematic image. The construction method of the enhanced cardiac magnetic resonance image generation model includes: Multiple patient image samples were acquired; each patient image sample included: plain scan cardiac magnetic resonance cine image and gadolinium-enhanced delayed-enhancement cardiac magnetic resonance image; Layer matching was performed on each scan layer of the plain cardiac magnetic resonance cine image and the gadolinium-delayed enhancement cardiac magnetic resonance image in each patient image sample to ensure that the paired scan layer images have a consistent cardiac cross-sectional position. For each matched patient image sample, plain cardiac magnetic resonance cine images and gadolinium-enhanced cardiac magnetic resonance images were subjected to image preprocessing based on the left ventricle as the center, with fixed resolution and pixel values ​​within a set range. Based on the contrast features of the left ventricular blood pool region extracted from the gadolinium-delayed enhancement cardiac magnetic resonance images in each patient image sample, contrast classification is performed on each patient image sample after image preprocessing to construct a patient image sample training set. A model for generating enhanced cardiac magnetic resonance images is trained based on the patient image sample training set. The layer matching of each scan layer in the plain cardiac magnetic resonance MRI cine image and the gadolinium-enhanced cardiac magnetic resonance image in each patient image sample includes: screening each patient image sample to remove plain cardiac magnetic resonance MRI cine images and corresponding gadolinium-enhanced cardiac magnetic resonance images with significant differences in imaging axis direction; calculating the projection value from the origin of the coordinates of each scan layer image of the plain cardiac magnetic resonance MRI cine image to the imaging axis of the corresponding gadolinium-enhanced cardiac magnetic resonance image in each screened patient image sample to obtain the two scan layer images with the closest position in the plain cardiac magnetic resonance MRI cine image and the corresponding gadolinium-enhanced cardiac magnetic resonance image; calculating the affine transformation relationship between the two scan layer images with the closest position in the obtained plain cardiac magnetic resonance MRI cine image and the corresponding gadolinium-enhanced cardiac magnetic resonance image for rigid registration of the two scan images; and performing non-rigid registration on the two scan images after rigid registration. The method of training the enhanced cardiac magnetic resonance imaging (MRI) image generation model based on the patient image sample training set includes: training a deep adversarial network based on the patient image sample training set to obtain a primary enhanced cardiac MRI image generation model, and generating a corresponding enhanced cardiac MRI image based on the plain scan cardiac MRI cine image of each sample in the patient image sample training set; optimizing the primary enhanced cardiac MRI image generation model based on the enhanced cardiac MRI images of each sample to obtain the final enhanced cardiac MRI image generation model. The step of generating corresponding enhanced cardiac magnetic resonance images based on the plain scan cardiac magnetic resonance cinema images of each sample in the patient image sample training set includes: constructing a feature map containing spatiotemporal information based on the dynamic, texture, and semantic feature information of the target scan layer and the images of the adjacent upper and lower scan layers of the plain scan cardiac magnetic resonance cinema images in the patient image sample training set; directionally adjusting the feature map by specifying a contrast category; selecting a frame image from the target scan layer of the plain scan cardiac magnetic resonance cinema images as a transfer source image by matching the triggered delayed imaging time based on the directionally adjusted feature map; wherein the cardiac phase state of the image frame is most similar to the cardiac phase state of the corresponding gadolinium-delayed enhanced cardiac magnetic resonance image; and using an image transfer method to transfer the selected image frame to generate a corresponding enhanced cardiac magnetic resonance image with a specified contrast category.

2. The method for generating enhanced cardiac magnetic resonance images according to claim 1, characterized in that, The preprocessing of plain cardiac magnetic resonance imaging (MRI) cine images and gadolinium-enhanced cardiac MRI images in each matched patient image sample, based on images centered on the left ventricle and with fixed resolution and pixel values ​​within a set range, includes: For each matched patient image sample, the plain cardiac magnetic resonance cinematic image and the corresponding gadolinium-enhanced delayed cardiac magnetic resonance image were resampled at a fixed resolution for each scan layer. Based on a pre-trained cardiac localization network, the resampled images of each scan layer are automatically localized and cropped to preserve the region of interest centered on the left ventricle. The cropped scan layer images are normalized so that all pixel values ​​fall within a set range.

3. The method for generating enhanced cardiac magnetic resonance images according to claim 1, characterized in that, The method for optimizing the primary enhanced cardiac magnetic resonance imaging (MRI) image generation model based on enhanced cardiac MRI images of each sample includes: Contrast augmentation was performed on the gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images of each sample in the patient image sample training set to generate augmented gadolinium-enhanced delayed-enhancement cardiac magnetic resonance images with different contrasts. The enhanced gadolinium-enhanced cardiac magnetic resonance image corresponding to each sample, the original gadolinium-enhanced cardiac magnetic resonance image, and the generated enhanced cardiac magnetic resonance image are compared. An iterative optimization algorithm is used to optimize the primary enhanced cardiac magnetic resonance image generation model to obtain the final enhanced cardiac magnetic resonance image generation model.

4. The method for generating enhanced cardiac magnetic resonance images according to claim 1, characterized in that, The step of generating enhanced cardiac magnetic resonance images with a specified contrast category based on the enhanced cardiac magnetic resonance image generation model includes: The acquired plain scan cardiac magnetic resonance cine images are preprocessed based on the left ventricle as the center, with a fixed resolution and pixel values ​​within a set range. By specifying the contrast category and ECG gating trigger time, a corresponding enhanced cardiac magnetic resonance image is generated based on the preprocessed plain scan cardiac magnetic resonance cinematic image.

5. A system for generating enhanced cardiac magnetic resonance images, characterized in that, include: The enhanced cardiac magnetic resonance image generation model construction module is used to perform scan layer matching, image preprocessing and contrast classification on plain scan cardiac magnetic resonance cinema images and corresponding gadolinium-delayed enhanced cardiac magnetic resonance images from multiple patient image samples to construct a patient image sample training set, and to train the enhanced cardiac magnetic resonance image generation model based on the patient image sample training set. The enhanced cardiac magnetic resonance image generation model prediction module is connected to the enhanced cardiac magnetic resonance image generation model training module. It is used to generate corresponding enhanced cardiac magnetic resonance images from the pre-processed plain scan cardiac magnetic resonance cinema images based on the constructed enhanced cardiac magnetic resonance image generation model by specifying the contrast category and the ECG gating trigger time. The layer matching process for each scan layer of the plain cardiac MRI cine image and the gadolinium-enhanced delayed cardiac MRI image in each patient image sample includes: screening each patient image sample to remove plain cardiac MRI cine images and corresponding gadolinium-enhanced delayed cardiac MRI images with significant differences in imaging axis direction; calculating the projection value from the origin of the coordinates of each scan layer of the plain cardiac MRI cine image to the imaging axis of the corresponding gadolinium-enhanced delayed cardiac MRI image in each screened patient image sample to obtain the two scan layer images with the closest position in the plain cardiac MRI cine image and the corresponding gadolinium-enhanced delayed cardiac MRI image; calculating the affine transformation relationship between the two scan layer images with the closest position in the obtained plain cardiac MRI cine image and the corresponding gadolinium-enhanced delayed cardiac MRI image for rigid registration of the two scan images; and performing non-rigid registration on the two scan images after rigid registration. The method of training the enhanced cardiac magnetic resonance imaging (MRI) image generation model based on the patient image sample training set includes: training a deep adversarial network based on the patient image sample training set to obtain a primary enhanced cardiac MRI image generation model, and generating a corresponding enhanced cardiac MRI image based on the plain scan cardiac MRI cine image of each sample in the patient image sample training set; optimizing the primary enhanced cardiac MRI image generation model based on the enhanced cardiac MRI images of each sample to obtain the final enhanced cardiac MRI image generation model. The step of generating corresponding enhanced cardiac magnetic resonance images based on the plain scan cardiac magnetic resonance cinema images of each sample in the patient image sample training set includes: constructing a feature map containing spatiotemporal information based on the dynamic, texture, and semantic feature information of the target scan layer and the images of the adjacent upper and lower scan layers of the plain scan cardiac magnetic resonance cinema images in the patient image sample training set; directionally adjusting the feature map by specifying a contrast category; selecting a frame image from the target scan layer of the plain scan cardiac magnetic resonance cinema images as a transfer source image by matching the triggered delayed imaging time based on the directionally adjusted feature map; wherein the cardiac phase state of the image frame is most similar to the cardiac phase state of the corresponding gadolinium-delayed enhanced cardiac magnetic resonance image; and using an image transfer method to transfer the selected image frame to generate a corresponding enhanced cardiac magnetic resonance image with a specified contrast category.

6. An electronic terminal, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory to cause the terminal to perform the enhanced cardiac magnetic resonance image generation method as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the enhanced cardiac magnetic resonance image generation method according to any one of claims 1 to 4.