Systems and methods for processing real-time cardiac MRI images

Abstract

Systems and methods for processing real-time cardiac MRI images are described that can continuously capture real-time cardiac MRI images across multiple cardiac phases and multiple slice images. Machine learning based techniques can be used to determine spatial (e.g., slice images and / or views) and temporal (e.g., cardiac cycles and / or cardiac phases) characteristics of the cardiac images so that the images can be arranged into groups based on the spatial and temporal characteristics of the images and requirements of a cardiac analysis task. Different groups of cardiac MRI images can also be aligned with each other based on timestamps of the images and / or by synthesizing additional images to fill in gaps.

Description

Technical Field

[0001] This application relates to medical image processing methods, particularly the processing of cardiac magnetic resonance images. Background Technology

[0002] Cardiac magnetic resonance imaging (CMR) is an important medical imaging tool for the detection and treatment of heart disease. Conventional CMR techniques typically require patients to hold their breath during the imaging process to minimize the effects of respiratory motion and may also require an electrocardiogram (ECG) to determine the cardiac phase of individual CMR images and / or combine data from multiple heartbeats to form a single synthetic cardiac systolic cycle. In recent years, alternative magnetic resonance imaging (MRI) techniques known as real-time CMR have been increasingly adopted due to their faster and more flexible operating modes. However, with real-time CMR, MRI signals (e.g., k-space data) can be acquired continuously (e.g., rather than always at the beginning of a specific cardiac phase or slice-by-slice images), thus presenting a challenge in determining the spatial and / or temporal alignment of the acquired images. Summary of the Invention

[0003] This document describes systems, methods, and apparatuses associated with real-time cardiac MRI image processing. According to embodiments of this disclosure, an apparatus capable of performing a real-time MRI image processing task may include at least one processor configured to: acquire a plurality of medical images of the heart and automatically determine a slice image and cardiac phase associated with each of the plurality of medical images based on one or more machine learning (ML) image recognition models. The plurality of medical images may be captured based on real-time MRI technology and may span multiple cardiac phases and multiple slice images of the heart. For example, the plurality of medical images may include a first medical image of the heart captured sequentially with a second medical image of the heart, wherein the first and second medical images may be associated with corresponding cardiac phases and slice images, and wherein the first and second medical images may be different from each other with respect to at least one of the cardiac phases or slice images associated with the first and second medical images. Based at least on the automatically determined slice image and cardiac phase information for each of the plurality of medical images and the requirements of the cardiac analysis task, the processor may also be configured to: select a first set of medical images from the plurality of medical images and provide the first set of medical images for the cardiac analysis task.

[0004] In the example, at least one processor of the device may also be configured to: automatically determine a view associated with each of a plurality of medical images based on one or more ML image recognition models, and further select a first set of images based on the view associated with each of the plurality of medical images. Such views may include, for example, a short-axis view of the heart, a two-chamber long-axis view, a three-chamber long-axis view, or a four-chamber long-axis view.

[0005] In the example, at least one processor of the device may also be configured to select a second set of medical images from a plurality of medical images, based at least on the requirements of the cardiac analysis task and the slice images and cardiac phases associated with each of the multiple medical images. The second set of medical images may be associated with cardiac cycles different from those of the first set of medical images described above, and the second set of medical images may be misaligned with the first set of medical images with respect to one or more time points. In this case, at least one processor of the device may be configured to generate one or more additional medical images of the heart for the second set of medical images, and to add the one or more additional medical images to the second set of medical images such that the second set of medical images is aligned with the first set of medical images with respect to one or more time points. The one or more additional medical images may be determined, for example, based on the corresponding timestamps of the medical images included in the first and second sets. The one or more additional medical images may be generated, for example, based on interpolation techniques or machine learning image synthesis models.

[0006] In the example, at least one processor of the device may also be configured to register a first medical image in a first set of medical images with a second medical image in the first set of medical images, wherein the registration may compensate for respiratory movements associated with the first or second medical image. In the example, at least one processor may also be configured to perform the aforementioned cardiac analysis task based on the first set of medical images. Attached Figure Description

[0007] The examples disclosed herein can be understood in more detail from the following description, which is given by way of example in conjunction with the accompanying drawings.

[0008] Figure 1 This is a simplified diagram illustrating an example of grouping real-time CMR images based on the spatial and temporal characteristics of images automatically determined using one or more machine learning (ML) image recognition models, according to some embodiments described herein.

[0009] Figure 2 This is a simplified diagram illustrating examples of artificial neural networks that can be used to implement and / or learn image classification models according to some embodiments described herein.

[0010] Figure 3This is a simplified diagram illustrating examples of artificial neural networks that can be used to implement and / or learn segmentation models according to some embodiments described herein.

[0011] Figure 4 This is a simplified diagram illustrating examples of artificial neural networks that can be used to implement and / or learn image registration models according to some embodiments described herein.

[0012] Figure 5 This is a flowchart illustrating an example method for training a neural network to perform one or more tasks described with respect to some embodiments provided herein.

[0013] Figure 6 This is a simplified block diagram illustrating an example system or device for performing one or more tasks described with respect to some embodiments provided herein. Detailed Implementation

[0014] The present disclosure is illustrated by way of example and not limitation in the accompanying drawings. A detailed description of illustrative embodiments will now be described with reference to the various drawings. Although detailed examples of possible implementations are provided in this specification, it should be noted that these details are intended to be exemplary and in no way intended to limit the scope of this application.

[0015] Figure 1An example is illustrated of grouping real-time CMR images based on the spatial and temporal characteristics of images automatically determined using one or more machine learning (ML) image recognition models. CMR image 102 (e.g., which may be obtained in Medical Digital Imaging and Communication (DICOM) format or other standard image formats such as JPEG) can be captured using real-time MRI techniques capable of continuously acquiring MRI signals (e.g., referred to herein as k-space data) and reconstructing them into CMR image 102. Using real-time MRI techniques, image capture can begin at any point in the cardiac cycle and can end at any point in the cardiac cycle or subsequent cardiac cycles. The image capture operation may not be accompanied by physiological information (e.g., such as information provided by ECG) or a patient's request to hold their breath during image capture. Furthermore, scans of the patient's heart can move between consecutive time points from one slice image (e.g., corresponding to an orthogonal 2D plane along the cardiac axis) to another slice image, rather than remaining at one slice image location for an extended period of time, as is the case in conventional or retrospective cinema CMR procedures (e.g., retrospective cinema CMR can capture a single slice image at a time). Therefore, CMR image 102 can span multiple cardiac phases (e.g., systole, diastole, etc.) and multiple slice images (e.g., along the short and long axes of the heart), and can be organized without regard to temporal and / or spatial order (e.g., as in retrospective cinematic CMR). For example, while CMR image 1 may be captured at time 1 in slice image position 1 (e.g., during systole), consecutively captured images (e.g., CMR image 2) may be associated with different slice images (e.g., slice image position 2) and / or different cardiac phases (e.g., during diastole). The individual views represented by consecutive images can also differ from one another. For example, while CMR image 1 may represent a short-axis view of the heart, CMR image 2 may represent a two-chamber long-axis view of the heart, and CMR image 3 may represent a three-chamber or four-chamber long-axis view of the heart.

[0016] Machine learning techniques can be used to automatically determine the temporal and spatial characteristics of CMR images 102, and to group the CMR images 102 based at least on these characteristics and the requirements of the specific clinical task to be performed (e.g., T1 / T2 mapping, tissue characterization, medical anomaly detection, etc.). Temporal characteristics may include, for example, the cardiac phase of each CMR image 102, the sequential order of the CMR images within a certain slice, etc., while spatial characteristics may include, for example, the slice to which each CMR image 102 belongs, the view represented by each CMR image 102, etc. For example, as... Figure 1As shown, one or more ML image recognition models 104 can be trained and used at 106 to automatically determine slice images, views, and / or cardiac phases associated with each CMR image 102. The automatically determined information can then be used, along with requirements 108 associated with one or more cardiac analysis tasks 110, to arrange the CMR images 102 into appropriate groups (e.g., at 112) for cardiac analysis tasks 110. The terms “machine learning,” “machine-based,” “deep learning,” and “artificial intelligence” are used interchangeably herein, as are the terms “machine learning model,” “neural network,” and “deep neural network.”

[0017] In the example, one or more ML image recognition models 104 may include a first image classification model trained to classify CMR images 102 into classes or categories corresponding to different slice images or views of the heart. For example, the first image classification model may use a deep neural network (DNN) to learn and / or implement to classify the individual CMR images 102 into view categories such as short-axis view, 2-chamber long-axis view, 3-chamber long-axis view, 4-chamber long-axis view, etc. The CMR images 102 may be further classified to determine whether they belong to the same slice image or different slice images (e.g., a view may correspond to the angle at which the image was acquired, while a slice image may correspond to a cut along that angle). For example, CMR image 102 may include a real-time CMR sequence comprising images {img_0_0_sax,img_1_0_sax,...,img_0_1_sax,img_1_1_sax,...,img_0_0_2ch,img_1_0_2ch,...,img_0_0_4ch,...}, wherein the first and second values ​​in the flags may represent the time of image capture and the location of the slice image, respectively, and the last part of the flags may represent the view captured in each image (e.g., minor axis (sax), 2-cavity major axis (2ch), 3-cavity major axis (3ch), 4-cavity major axis (4ch), etc.). The DNN may be trained to learn features associated with each slice image and view through a training process, and subsequently automatically determine the slice image and / or view associated with a given image based on the learned features. The determined slice images and / or view information can then be used to arrange the CMR images 102 into different groups, including, for example, a first group {img_0_0_sax,img_1_0_sax...} that may correspond to slice images 0 in the short-axis view, a second group {img_0_1_sax,img_1_1_sax...} that may correspond to slice images 1 in the short-axis view, a third group {img_0_0_2ch,img_1_0_2ch...} that may correspond to slice images 0 in the long-axis view of the 2-cavity core, a fourth group {img_0_0_4ch...} that may correspond to slice images 0 in the long-axis view of the 4-cavity core, and so on. The DNN can use the CMR images 102 as input (e.g., only as input) to determine the slice image / view information of each image, or the DNN can use the CMR images 102 and other acquisition information (e.g., such as the absolute acquisition time and / or position included in the DICOM header) as input for determination. In the latter case, the first image classification model can utilize CMR images and acquired information together or sequentially.

[0018] In the example, one or more ML image recognition models 104 may include a second image classification model trained to determine the cardiac phase (e.g., within the cardiac cycle) associated with each CMR image 102. Such an ML model may also be learned and implemented using a DNN, which may be trained to learn features associated with various cardiac phases (e.g., end-diastolic (ED), end-systolic (ES), etc.) through a training process, and then automatically predict the cardiac phase depicted in a given image based on the learned features. For example, CMR image 102 may include images {img_0_0_sax, img_0.3_0_sax, img_0.6_0_sax, img_0.9_0_sax, img_1.2_0_sax, img_1.5_0_sax, img_1.8_0_sax, img_2.1_0_sax...} spanning one or more cardiac phases or cycles, where the first number in the label may represent the absolute acquisition time of the image (e.g., 0 may not necessarily correspond to the start of the cardiac cycle), and the second number in the label may represent the slice image to which the image may belong. Based on features learned through training, the DNN can classify img_0.3_0_sax and img_1.5_0_sax as belonging to ED, and img_0.9_0_sax and img_2.1_0_sax as belonging to ES. Then, based on the determination of these key cardiac phases, CMR images 102 can be grouped into {img_0_0_sax}, {img_0.3_0_sax,img_0.6_0_sax,img_0.9_0_sax,img_1.2_0_sax}, and {img_1.5_0_sax,img_1.8_0_sax,img_2.1_0_sax}, where each group can correspond to a cardiac cycle and can include images that begin at one end of an ED and end before the next ED. The remaining images can be assigned to these groups based on the detected key cardiac phases, as these images may have been captured sequentially in time to reflect the continuous motion of the heart. In the example, a timestamp or time position relative to the first image in a group (e.g., it could be an ED) can be assigned to each image in the group, allowing the images to be aligned with images from another group, as will be described in more detail below. The DNN can use (e.g., only) the CMR image 102 as input to determine the cardiac phase of the image, or the DNN can use the CMR image 102 and other acquisition information (e.g., such as the absolute acquisition time included in the DICOM header) as input for determination. In the example, additional ML-based processing can be performed to facilitate the detection of cardiac phase.For example, a segmentation network can be used to segment the heart depicted in CMR image 102 to obtain volumetric information (e.g., such as left ventricular (LV) volume) as depicted in the image. This volumetric information can then be used to facilitate the determination of cardiac phases, for example, since the ED and / or ES phases are strongly correlated with LV volume. The results of the segmentation operation can be reused in subsequent image analysis tasks (e.g., post-analysis tasks) without incurring additional costs.

[0019] Using automatically determined temporal (e.g., cardiac phase) and spatial (e.g., slice image / view) information of the CMR images 102 and requirements 108 associated with cardiac analysis task 110, at 112, the CMR images 102 can be grouped according to requirements 108 and the automatically determined information. For example, for some cardiac analysis tasks, it may be desirable to examine CMR images 102 grouped into different slice images, where each slice image may include images depicting cardiac motion, while for other analysis tasks, it may be desirable to group the CMR images 102 into different cardiac phases, where each phase may include images spanning multiple slice images and containing information about the entire heart.

[0020] In the example, when images are labeled with cardiac phase information and / or timestamps automatically determined in 106, all or a subset of CMR images 102 can be arranged into groups corresponding to the respective cardiac cycles. However, two or more of these image groups may not be time-aligned (e.g., for patients with heart conditions such as premature ventricular contractions, which may cause the timing of cardiac cycles to vary from one cycle to the next). For example, a first group corresponding to a first cardiac cycle may include three images with corresponding timestamps or time positions 1, 3, and 5 (e.g., relative to the first image in the first group), while a second group corresponding to a second cardiac cycle may include five images with corresponding timestamps or time positions 1, 2, 3, 4, and 5 (e.g., relative to the first image in the second group). For clinical applications that may require time alignment of CMR images across different cardiac cycles, multiple CMR images from a set can be merged (e.g., merged into a single image), or additional CMR images can be generated for a set (e.g., at 2 and 4 time locations in a first set) such that the images in that set can be aligned with images from another set (e.g., the second set mentioned above). Additional images can be generated using various interpolation techniques (e.g., linear interpolation techniques) based on existing images within and / or across different cardiac cycles (e.g., using corresponding timestamps determined for images associated with different cardiac cycles). Additional images can also be generated using neural networks trained for image synthesis, for example, by utilizing temporally related neighboring images.

[0021] In the example, one or more CMR images 102 (e.g., belonging to the same slice or cardiac phase) can be captured while the patient is in motion (e.g., respiratory motion). The effect of this motion on the CMR images (e.g., reflected by in-plane translation) can be compensated for using various image registration techniques, including, for example, image registration neural networks. Motion compensation operations can be combined with motion-related operations (e.g., such as motion estimation) in a post-analysis process, such that the results of the motion compensation operations can be reused during the post-analysis process without incurring additional computation or resource usage. In the example, respiratory motion can be removed without motion compensation operations, and through-plane translation (e.g., cross-slice motion correction) can be achieved by separating the CMR images into slice images and grouping multiple slice images together, for example, using the image alignment techniques described herein.

[0022] Figure 2 Examples of artificial neural networks (ANNs) are illustrated, which can be used to implement and / or learn image classification models, such as those described herein (e.g., for classifying CMR images into different slice images, views, and / or cardiac phases). This example is provided for illustrative purposes and not to limit the scope of this disclosure. Those skilled in the art will understand that other machine learning models, such as visual transformers, can also be used to achieve the objectives described herein. Figure 2 As shown, the ANN can be a convolutional neural network (CNN), which may include multiple layers, such as one or more convolutional layers 202, one or more pooling layers 204, and / or one or more fully connected layers 206. Each convolutional layer 202 may include multiple convolutional kernels or filters configured to extract features from the input image 208 (e.g., a movie image). Following the convolution operation may be batch normalization and / or linear (or non-linear) activation, and the features extracted by the convolutional layers may be downsampled by pooling layers and / or fully connected layers to reduce feature redundancy and / or size, thereby obtaining a representation of the downsampled features (e.g., in the form of feature vectors or feature maps). Based on the feature representation, classification prediction may be performed, for example, at the output of the ANN to indicate whether the input image 208 is a minor-axis image (e.g., a minor-axis slice image), a major-axis image (e.g., a major-axis slice image), a 2-cavity image, a 3-cavity image, or a 4-cavity image, etc. Classification can be indicated by labels (e.g., “minor axis image”, “major axis image”, “2-cavity image”, etc.) or numerical values ​​(e.g., 1 corresponds to the minor axis image, 2 corresponds to the major axis image, 3 corresponds to the 2-cavity image, etc.).

[0023] Figure 3An example of an artificial neural network (ANN) 302 is illustrated, which can be used to implement and / or learn image segmentation models, such as the image segmentation models described herein (e.g., for determining volumetric information of the heart based on CMR images in order to determine the cardiac phase associated with the CMR images). The ANN 302 can utilize an architecture including an encoder network and a decoder network. The encoder network can be configured to receive an input image 304, such as a movie image, extract features from the input image, and generate a representation of the features at the output (e.g., a low-resolution or low-dimensional representation). The encoder network can be a convolutional neural network with multiple layers configured to extract and downsample features from the input image 304. For example, the encoder network can include one or more convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Each convolutional layer can include multiple convolutional kernels or filters configured to extract specific features from the input image. Batch normalization and / or non-linear activation can follow the convolutional operation, and the features extracted by the convolutional layers (e.g., in the form of one or more feature maps) can be downsampled by pooling layers and / or fully connected layers to reduce feature redundancy and / or size. The feature representations generated by the encoder network can take various forms, including, for example, feature maps or feature vectors.

[0024] The decoder network of ANN 302 can be configured to receive representations generated by the encoder network, decode features of the input image 304 based on these representations, and generate masks 306 (e.g., pixel-level or voxel-level segmentation masks) for segmenting one or more objects (e.g., LV and / or RV of the heart, AHA cardiac segments, etc.) from the input image 302. The decoder network may also include multiple layers configured to perform upsampling and / or transposed convolution (e.g., deconvolution) operations on the feature representations generated by the encoder network and recover spatial details of the input image 304. For example, the decoder network may include one or more upper pooling layers and one or more convolutional layers. Through the upper pooling layers, the decoder network can upsample the feature representations generated by the encoder network (e.g., based on pooling indices stored by the encoder network). Then, the upsampled representations can be processed by the convolutional layers to generate one or more dense feature maps before applying batch normalization to one or more dense feature maps to obtain a high-dimensional representation of the input image 304. As described above, the output of the decoder network may include a segmentation mask for depicting one or more anatomical structures or regions from the input image 304. In the example, this segmentation mask can correspond to a multi-class pixel / voxel level probability map, where pixels or voxels belonging to each of the multiple classes are assigned a high probability value indicating the classification of the pixel / voxel.

[0025] Figure 4 An example is shown using an artificial neural network (ANN) 402 to register two CMR images I mov(e.g., source CMR images) and I fix (e.g., a target CMR image) as an example to compensate for motion (e.g., breathing movements) associated with the image. As shown, the neural network 402 can be configured to receive image I fix and I mov (For example, as input), image I mov From the moving image domain (e.g., with image I) mov (Associated) transformation to a fixed image domain (e.g., with image I) fix (Associated), and generate image I reg (For example, as image I) mov (spatial transformation version), this image I reg It can be similar to image I fix (For example, in I) fix with I reg There is a minimized dissimilarity between them (404). A neural network (402) can be trained to determine the method for processing image I. mov Transform into image I reg Multiple transformation parameters θ T As illustrated by the following equation:

[0026] I reg =I mov ((θx)) (1)

[0027] Where x can represent the coordinates in the moving image domain, θ(x) can represent the mapping of x to the fixed image domain, and I mov ((θx)) can represent one or more grid sampling operations (e.g., using sampler 406). θ can include parameters associated with an affine transformation model that allows translation, rotation, scaling, and / or skew of the input image. θ can also include parameters associated with a deformable field (e.g., a dense deformable field) that allows deformation of the input image. For example, θ can include stiffness parameters, B-spline control points, deformable parameters, etc.

[0028] Figure 5A flowchart illustrating an example process 500 for training a neural network (e.g., an ML model implemented by a neural network) to perform one or more tasks described herein. As shown, the training process 500 may include: at 502 initializing the execution parameters of the neural network (e.g., weights associated with the individual layers of the neural network), for example by sampling from a probability distribution or by replicating the parameters of another neural network with a similar structure. The training process 500 may also include at 504 processing the input training image (e.g., a movie image or tissue representation map) using the currently assigned parameters of the neural network, and at 506 making a prediction about the desired outcome (e.g., classification labels, segmentation masks, motion-compensated images, etc.). At 508, the prediction may be compared to a gold standard, and a loss associated with the prediction may be determined based on this comparison and a loss function. The loss function used for training may be selected based on the specific task in which the neural network is trained. For example, if the task involves classification or segmentation of the input image, or minimizing the difference between the original image and the distorted image (e.g., for image registration), a loss function based on the mean squared error between the prediction and the gold standard may be used.

[0029] At step 510, the loss calculated using one or more of the techniques described above can be used to determine whether one or more training termination criteria are met. For example, if the loss is below a threshold or if the change in loss between two training iterations is below a threshold, it can be determined that the training termination criteria are met. If the termination criteria are determined to be met at step 510, training can end; otherwise, at step 512, for example, before training returns to step 506, the currently assigned network parameters can be adjusted by backpropagating the gradient descent of the loss function through the network.

[0030] For the sake of simplicity, the training steps are depicted and described in a specific order herein. However, it should be understood that training operations can occur in various orders, simultaneously, and / or with other operations not presented or described herein. Furthermore, it should be noted that not all operations that may be included in the training method are depicted and described herein, and not all exemplified operations need to be performed.

[0031] The systems, methods, and / or apparatuses described herein may be implemented using one or more processors, one or more storage devices, and / or other suitable auxiliary devices (such as display devices, communication devices, input / output devices, etc.). Figure 6An example of a device 600 that can be configured to perform the tasks described herein is illustrated. As shown, device 600 may include at least one processor (e.g., one or more processors) 602, which may be a central processing unit (CPU), graphics processing unit (GPU), microcontroller, reduced instruction set computer (RISC) processor, application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), physical processing unit (PPU), digital signal processor (DSP), field-programmable gate array (FPGA), or any other circuitry or processor capable of performing the functions described herein. Device 600 may also include communication circuitry 604, memory 606, mass storage device 608, input device 610, and / or communication link 612 (e.g., communication bus) through which one or more components shown in the figures exchange information.

[0032] Communication circuitry 604 can be configured to send and receive information using one or more communication protocols (e.g., TCP / IP) and one or more communication networks, including local area networks (LANs), wide area networks (WANs), the Internet, and wireless data networks (e.g., Wi-Fi, 3G, 4G / LTE, or 5G networks). Memory 606 may include a storage medium (e.g., a non-transitory storage medium) configured to store machine-readable instructions that, when executed, cause processor 602 to perform one or more functions described herein. Examples of machine-readable media may include volatile or non-volatile memory, including but not limited to semiconductor memory (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)), flash memory, etc.). Mass storage device 608 may include one or more disks, such as one or more internal hard disks, one or more removable disks, one or more magneto-optical disks, one or more CD-ROMs or DVD-ROMs, etc., on which instructions and / or data may be stored for operation of processor 602. Input device 610 may include a keyboard, mouse, voice-controlled input device, touch-sensitive input device (e.g., touch screen), etc., for receiving user input from device 600.

[0033] It should be noted that device 600 can operate as a standalone device or can be connected to other computing devices (e.g., networked or clustered) to perform the tasks described herein. And even in Figure 6 Only one example of each component is shown in the figure, and those skilled in the art will understand that device 600 may include multiple instances of one or more components shown in the figure.

[0034] Although this disclosure has been described according to certain embodiments and generally associated methods, changes and variations of the embodiments and methods will be apparent to those skilled in the art. Therefore, the above description of exemplary embodiments does not limit this disclosure. Other changes, substitutions, and modifications are possible without departing from the spirit and scope of this disclosure. Furthermore, unless specifically stated otherwise, discussions using terms such as “analyze,” “determine,” “enable,” “identify,” and “modify” refer to the actions and processes of a computer system or similar electronic computing device that manipulate and transform data representing physical (e.g., electronic) quantities within the registers and memories of the computer system into other data representing physical quantities within the computer system's memory or other such information storage, transmission, or display devices.

[0035] It should be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will become apparent to those skilled in the art after reading and understanding the above description.

Claims

1. A method for processing medical images, the method comprising: Obtain multiple medical images of the heart; The slice image and cardiac phase associated with each of the plurality of medical images are determined based on one or more machine learning (ML) image recognition models. A first set of medical images is selected from the plurality of medical images based at least on the slice images and cardiac phases associated with each of the medical images in the plurality of medical images and the requirements of the cardiac analysis task; as well as The first set of medical images is provided for performing the cardiac analysis task.

2. The method according to claim 1, wherein, The multiple medical images were captured using real-time magnetic resonance imaging (MRI) technology and span multiple cardiac phases and multiple slice images of the heart.

3. The method according to claim 2, wherein, The plurality of medical images includes a first medical image of the heart captured sequentially with a second medical image of the heart, the first medical image and the second medical image being associated with corresponding cardiac phases and slice images, and wherein the first medical image and the second medical image are different from each other with respect to at least one of the cardiac phases or slice images associated with the first medical image and the second medical image.

4. The method according to claim 1, further comprising: The view associated with each of the plurality of medical images is automatically determined based on the one or more ML image recognition models, and the first group of medical images is selected based on the view associated with each of the plurality of medical images, wherein the view includes a 2-chamber view, a 3-chamber view or a 4-chamber view of the heart.

5. The method according to claim 1, wherein, The first set of medical images is associated with a first cardiac cycle, and the method further includes: At least based on the requirements of the cardiac analysis task and the slice images and cardiac phases associated with each of the plurality of medical images, a second set of medical images is selected from the plurality of medical images, wherein the second set of medical images is associated with a second cardiac cycle and is not aligned with the first set of medical images with respect to one or more time points; Generate one or more additional medical images of the heart based on the second set of medical images; and Add the one or more additional medical images to the second set of medical images such that the second set of medical images is aligned with the first set of medical images with respect to the one or more time points.

6. The method according to claim 5, further comprising: Determine the corresponding timestamps of the medical images included in the first group of medical images and the second group of medical images, wherein the one or more additional medical images are generated for the second group of medical images based at least on the determined timestamps.

7. The method according to claim 5, wherein, The one or more additional medical images are generated using interpolation techniques or machine learning image synthesis models.

8. The method according to claim 1, further comprising: The first medical image in the first set of medical images is registered with the second medical image in the first set of medical images, wherein the registration compensates for respiratory movements associated with the first medical image or the second medical image.

9. The method according to claim 1, further comprising: The cardiac analysis task is performed based on the first set of medical images.

10. A computer program product comprising instructions that, when run on a computer, cause the computer to perform the method as described in any one of claims 1-9.

Images