System and method for generating a bull's eye plot

Abstract

Bull's eye plots can be generated based on cardiac magnetic resonance imaging (CMRI) to facilitate diagnosis and treatment of heart disease. Systems, methods, and devices associated with bull's eye plot generation are described herein. Multiple myocardial segments can be obtained to construct a bull's eye plot based on landmark points detected in short-axis and long-axis magnetic resonance (MR) slice images of the heart and by sequentially arranging the short-axis MR slice images according to the order in which the slice images were generated during the CMRI. The sequential order of the short-axis MR slice images can be determined using the projected positions of the short-axis MR slice images on the long-axis MR slice images and the corresponding distances of the projected positions to the landmark points of the long-axis MR slice images. An artificial neural network can be used to identify myocardium and / or landmark points in the short-axis and / or long-axis MR slice images.

Description

Technical Field

[0001] The present application relates to the field of magnetic resonance imaging. Background Art

[0002] Cardiac magnetic resonance imaging (CMRI) is a standard imaging modality for non-invasive cardiac diagnosis. Using CMRI, high-resolution and multi-slice anatomical images of the human heart can be acquired during multiple cardiac cycles during different scans or examination processes. The CMRI data can then be synthesized and presented to visualize regions of the human heart and provide an intuitive overview of cardiac conditions, including, for example, myocardial mass, myocardial strain, wall thickness, etc. An example technique for presenting CMRI data is by constructing a bull's eye diagram in which the myocardium is divided into multiple standard segments, and determining quantitative and / or qualitative information such as myocardial strain for each segment. Diagnosis and / or treatment can then be performed by referring to the standard segments of the myocardium in the bull's eye diagram and the quantitative and / or qualitative information associated with each segment. Because the bull's eye diagram plays an important role in clinical decision-making, it is highly desirable to ensure that the standard myocardial segments described herein are accurately determined based on CMRI imaging, and to construct the bull's eye diagram to truly reflect the anatomical structure of the human heart. Summary of the invention

[0003] The present invention describes systems, methods, and devices associated with generating a bull's eye diagram based on CMRI imaging. A bull's eye diagram generating device described herein may include one or more processors configured to obtain a plurality of first magnetic resonance (MR) slice images based on a first cardiac magnetic resonance imaging (CMRI) scan performed along the short axis of a human heart, and to obtain a second MR slice image based on a second CMRI scan performed along the long axis of the human heart. The one or more processors may also be configured to determine one or more first landmark points associated with the human heart and a second landmark point associated with the human heart, and to facilitate the generation of a bull's eye diagram using the first landmark point and the second landmark point. For example, the one or more first landmark points may indicate the center of the left ventricle (LV) of the human heart and / or the intersection of the LV and the right ventricle (RV) of the human heart, the second landmark point may indicate the apex of the myocardium, and the one or more processors may be configured to determine (e.g., using metadata associated with the CMRI scan) the corresponding projection positions of the first MR slice image on the second MR slice image. Based on the corresponding distances of these projection positions to the second landmark points, the one or more processors may, for example, sequentially arrange the first MR slice images according to the original scanning order of the first MR slice images. The one or more processors may then determine a plurality of myocardial segments to be included in the bull's eye map based on the sequentially arranged first MR slice images and the one or more first landmark points, and generate the bull's eye map using the plurality of myocardial segments.

[0004] In an embodiment, one or more processors of the bull's eye image generating device may be configured to use an artificial neural network to segment at least one slice image in the first MR slice image to identify the LV, RV and myocardium. The one or more processors of the bull's eye image generating device may also be configured to use an artificial neural network to determine the first landmark point and / or the second landmark point. In an embodiment, the first MR slice images may be arranged sequentially from the base slice image to the apex slice image based on the corresponding distance from the projection position of the first MR slice image to the apex of the myocardium, wherein the base slice image may correspond to the projection position with the longest distance from the apex, and the apex slice image may correspond to the projection position with the shortest distance from the apex. In an embodiment, the one or more processors of the bull's eye image generating device may also be configured to determine the direction of the first CMRI scan based on the corresponding distance from the projection position of the first MR slice image to the apex of the myocardium. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Examples disclosed herein may be understood in more detail from the following description given by way of example in conjunction with the accompanying drawings.

[0006] Figure 1A and Figure 1B is a diagram illustrating an example multi-segment bull's eye plot of the myocardium and an example manner of obtaining the segments of the bull's eye plot.

[0007] Figure 2 is a diagram illustrating an example of segmenting a cardiac image and determining a plurality of myocardial segments that may be included in a bull's eye diagram.

[0008] Figure 3 is a block diagram illustrating an example of segmenting a cardiac image to identify myocardium in the image.

[0009] Figure 4 is a diagram illustrating an example of determining a plurality of landmark points in a segmentation mask and obtaining a plurality of myocardial segments based on the landmark points.

[0010] Figure 5 is a diagram illustrating an example of arranging short-axis MR slice images based on the sequential order of corresponding long-axis MR images.

[0011] Figure 6 is a flow chart illustrating example operations of the bull's-eye plot generation apparatus described herein.

[0012] Figure 7 is a flow chart illustrating an example process for training a neural network, which process may be implemented by the bull's-eye plot generation apparatus described herein.

[0013] Figure 8 is a block diagram illustrating example components of a bulls-eye plot generation apparatus described herein. DETAILED DESCRIPTION

[0014] The present disclosure is illustrated by way of example and not limitation in the figures of the accompanying drawings.

[0015] Figure 1A An example bull's eye diagram 100 of a human heart is illustrated, and Figure 1B The manner of dividing the myocardium into the segments included in the bull's eye diagram 100 is illustrated. As shown, the bull's eye diagram 100 may include seventeen segments. Sixteen of the seventeen segments may each belong to the base portion of the myocardium, the middle (or central) portion of the myocardium, or the apex portion of the myocardium. For example, the base portion may include six segments labeled 1-6, respectively, the middle portion may also include six segments labeled 7-12, respectively, and the apex portion may include four segments labeled 13-16, respectively. In addition, the bull's eye diagram 100 may also include a 17th segment that may correspond to the apex of the heart, such as the apex of the myocardium.

[0016] The bull's eye plot 100 may be generated based on images of the heart generated during different phases of the cardiac cycle, including, for example, the end-diastolic (ED) phase and / or the end-systolic (ES) phase. Figure 2 An example technique that can be used by a bull's eye diagram generation device described herein to construct a bull's eye diagram (e.g., bull's eye diagram 100) based on an image of the heart is illustrated. The image can be obtained from one or more magnetic resonance imaging (MRI) scans of the heart, and can include a plurality of magnetic resonance (MR) slice images derived from the MRI scans. For example, the MR slice images can include a plurality of first slice images derived from a first MRI scan along the short axis of a human heart. The MR slice images can also include a plurality of second MR slice images derived from a second MRI scan along the long axis of a human heart. When described herein, the long axis of the heart can be an axis that aligns the base of the heart and the apex of the heart, and the long axis of the heart can be an axis perpendicular to the short axis of the heart.

[0017] like Figure 2As shown, the bull's eye image generating device can obtain multiple first MR slice images, which can include MR slice images 202 taken along the short axis of the heart. The MR slice image 202 can include multiple MRI images of the heart taken within a period of time (e.g., one or more cardiac cycles) (e.g., in the form of a movie). Each MRI image in the MR slice image 202 can include a visual representation of all or part of the left ventricle (LV), right ventricle (RV), myocardium, left atrium and / or right atrium. Based on these images and the visual representation of the heart contained therein, the bull's eye image generating device can identify (e.g., segment) LV 204, RV 205 and / or myocardium 206, and further divide the myocardium 206 into multiple segments, which can then be used to construct the bull's eye image shown in Figure 1. For example, the MR slice image 202 may include a visual representation of a heart base portion of the myocardium 206, and the bull's eye diagram generation device may be configured to segment the myocardium 206 based on the image included in the MR slice image 202, and further divide the myocardium 206 into six segments 208 corresponding to segments 1-6 of the bull's eye diagram in FIG. 1 (e.g., in FIG. 1 ). Figure 2 Similarly, the bull's eye diagram generating device can also be configured to segment and divide the middle and apical portions of the myocardium 206 based on the MR slice images representing these portions of the myocardium 206 (for example, into six segments and four segments, respectively).

[0018] The bull's eye diagram generation device may include an artificial neural network trained to perform the segmentation tasks described herein. Figure 3An example segmentation neural network 302 is shown that can be implemented by a bull's eye map generation device to segment myocardium 304 from an image 306 of a human heart (e.g., to depict myocardium 304 in image 306). As described herein, image 306 can be an MRI image of a heart included in an MR slice image (e.g., generated by a CMRI). Image 306 can include a plurality of pixels having corresponding features (e.g., varying brightness, contrast, intensity, gradient, etc.). Segmentation neural network 302 can be a convolutional neural network (CNN), such as a deep convolutional neural network, which has been trained to identify one or more portions (e.g., one or more pixels) of image 306 as belonging to myocardium 304 and indicating myocardium (e.g., the location and / or boundaries of myocardium), for example, via a segmentation map or segmentation mask. Segmentation neural network 302 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 a plurality of kernels or filters configured to identify specific features (e.g., key points) or patterns in image 306. The kernels or filters can be associated with corresponding weights that can be learned via a training process. The convolution operation of each convolutional layer may be followed by batch normalization and / or activation (e.g., using a rectified linear unit (ReLU) function) before the output of one convolutional layer (e.g., in the form of a feature map) is provided as an input to the next convolutional layer. The feature map generated by the convolutional layer may be downsampled by one or more pooling layers (e.g., using a 2×2 window and a stride of 2) of the segmentation neural network 302, for example, to reduce the redundancy and / or size of features identified in the image 306 (e.g., by a factor of 2).

[0019] Subsequently, the downsampled features can be upsampled by the segmentation neural network 302, for example, via a series of transposed convolution operations (e.g., using a 3×3 transposed convolution kernel with a stride of 2) to recover the details associated with the identified features. One or more dense feature maps can be derived based on these operations, which can indicate visual features of various regions or pixels of the image 306. Based on these visual features, a subset of regions or pixels of the image 306 can be classified (e.g., via a fully connected layer of the segmentation neural network 302) as belonging to the myocardium 304, and a segmentation mask 308 can be generated to indicate the classification and / or segmentation. The segmentation neural network 302 can learn visual features of the myocardium from a training data set that can include a large number of MRI images of the human heart, and the parameters of the segmentation neural network 302 (e.g., weights described herein) can be acquired based on gradient descent associated with a loss function that represents the difference between the segmentation estimated by the neural network and the gold standard for segmentation (e.g., annotated regions of the myocardium). The training of the segmentation neural network 302 will be described in more detail below. Examples of neural networks that can be trained to perform segmentation tasks can be found in commonly assigned U.S. patent application Ser. No. 16 / 905,115, filed on Jun. 18, 2020, and entitled “Systems and Methods for Image Segmentation,” and U.S. patent application Ser. No. 17 / 014,594, filed on Sept. 8, 2020, and entitled “Hierarchical Systems and Methods for Image Segmentation,” the disclosures of which are hereby incorporated by reference in their entirety.

[0020] although Figure 3 306, but the segmentation neural network 302 can also be trained to segment (e.g., simultaneously with the segmentation of the myocardium) other parts of the human heart, such as the LV and RV, based on the image 306. Alternatively or additionally, the bull's eye image generation device described herein may implement additional neural networks configured to segment those other parts of the heart (e.g., the LV or RV). These additional neural networks can have a similar structure to the segmentation neural network 302, but can be trained (e.g., using a different training data set than the segmentation neural network 302) to specifically identify other parts of the heart.

[0021] The bull's eye image generation device described in this article can also be configured to determine one or more first landmark points associated with the human heart based on multiple first MR slice images, and determine one or more second landmark points associated with the human heart based on multiple second MR slice images, and use the first landmark points and the second landmark points to facilitate the generation of the bull's eye image. Figure 4An example of dividing the myocardium using landmark points 402a, 402b, and 402c to obtain segments for a bull's eye plot is shown. In this example, landmark points 402a and 402b may represent points where the RV intersects the LV (e.g., at the anterior and inferior LV myocardium), and landmark point 402c may represent the center of the LV. Landmark points 402a and 402b may divide the myocardium into two portions 404 and 406, wherein portion 404 may be shorter than portion 406. Thus, in order to obtain segments of the basal portion and the middle portion of the myocardium for a bull's eye plot, the bull's eye plot generation device may be configured to divide the shorter portion 404 into (e.g., equally) two segments and the longer portion 406 into four segments (e.g., corresponding to Figure 2 To obtain segments of the apical portion of the myocardium for a bull's eye view, the bull's eye view generating device may be configured to treat the shorter portion 404 as one myocardial segment and to divide the longer portion 406 into (e.g., equally) three segments (e.g., corresponding to segments 13-16 shown in FIG. 1 ).

[0022] The bull's eye diagram generating device may be configured to further utilize the landmark point 402c to perform the segmentation of the myocardium. For example, the bull's eye diagram generating device may be configured to draw a line 408 (e.g., a dashed line) between the landmark point 402c and one of the landmark points 402a and 402b, and rotate the line 408 around the myocardium with equal angular steps to obtain segments for the bull's eye diagram. If the myocardium is to be divided into six segments (e.g., for the bottom portion and the middle portion), such angular steps may be, for example, 60 degrees, and if the myocardium is to be divided into four segments (e.g., for the apex portion), such angular steps may be 90 degrees.

[0023] In an example embodiment, a bull's eye map generation device may be configured to use a landmark detection neural network to identify one or more landmark points described herein (e.g., such as landmark points 402a and 402b). The landmark detection neural network may include a convolutional neural network such as a deep convolutional neural network. The network may include one or more convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Each convolutional layer may include multiple kernels or filters configured to identify specific features (e.g., key points) or patterns in an input image. The kernel or filter may be associated with a corresponding weight that can be learned via a training process. Before the output of the convolutional layer (e.g., in the form of a feature map) is provided to the next convolutional layer and used to extract higher-level or higher-order features from the input image, the convolution operation performed via each convolutional layer may be followed by batch normalization and / or activation (e.g., using a rectified linear unit (ReLU) function). The feature maps generated by the convolutional layers can be downsampled by one or more pooling layers (e.g., using a 2×2 window and a stride of 2) of the landmark detection neural network, for example, to reduce the redundancy and / or size of the features identified in the input image (e.g., by a factor of 2). Subsequently, the downsampled features can be upsampled by the network, for example, via a series of transposed convolution operations (e.g., using a 3×3 transposed convolution kernel with a stride of 2) to recover the details associated with the identified features. One or more dense feature maps can be derived based on these operations to indicate visual features of various regions or pixels of the input image. Based on these visual features, the landmark detection neural network can identify regions or pixels in the input image corresponding to landmark points of interest (e.g., based on matching visual features) and determine the corresponding positions of the landmark points accordingly. The landmark detection neural network can learn the visual features of the landmark points from a training dataset that can include a large number of MRI images of the human heart, and the parameters of the network (e.g., weights described herein) can be acquired based on gradient descent associated with a loss function that represents the difference between the landmark positions estimated by the network and a gold standard for the positions (e.g., the annotated positions of the landmark points). The training of the landmark detection neural network will be described in more detail below.

[0024] In an example embodiment, the bull's eye image generation device can be configured to identify one or more landmark points described herein based on a segmentation mask or segmentation map that has been generated for a portion of a human heart. For example, the bull's eye image generation device can use a segmentation neural network (e.g., such as the segmentation neural network 302 described herein) to obtain a segmentation mask for the LV, and determine the center of the LV (e.g., landmark point 402c) based on pixels that have been identified as part of the LV. For example, the bull's eye image generation device can determine the center of the LV based on an average (e.g., a weighted average) of coordinates of pixels located on the endocardial boundary or the epicardial boundary.

[0025] In order to generate a bull's eye image based on a CMRI scan, the bull's eye image generation device may need to sequentially determine the direction (e.g., orientation) of the CMRI scan and / or the MR slice images (e.g., short-axis MR slice images described herein) from which the MR slice images are obtained (e.g., according to the order in which the MR slice images are generated during the scan or inverse to the order in which the MR slice images are generated during the scan). However, in some cases, the direction of the CMRI scan may not be known to the bull's eye image generation device, and / or the MR slice images may not be sequentially organized (e.g., according to the scanning order of the MR slice images or inverse to the scanning order of the MR slice images). Therefore, in response to obtaining a plurality of MR slice images (e.g., short-axis MR slice images) based on the CMRI scan, the bull's eye image generation device may be configured to sequentially arrange (e.g., rearrange) the MR slice images, for example, in an order that conforms to the scanning order of the MR slice images. The bull's eye image generation device may also be configured to determine the direction (e.g., orientation) of the CMRI scan based on the rearranged MR slice images. The bull's eye diagram generating device can accomplish these tasks by determining the corresponding projection positions of the short-axis MR slice images on the long-axis MR slice images, and sorting the short-axis slice images based on the order of the corresponding distances of the projection positions to the landmark points on the long-axis MR slice images.

[0026] Figure 5 An example of arranging a plurality of short-axis MR slice images and / or determining the direction (e.g., orientation) of a CMRI scan based on a projection position 502 of the short-axis MR slice images on a long-axis MR slice image 504 and / or a landmark point 506 is illustrated. As described herein, a plurality of short-axis MR slice images may be generated according to a particular scan order (e.g., from the heart base to the heart apex or vice versa) during a CMRI scan. However, when provided to a bull's eye diagram generating apparatus as described herein, the short-axis MR slice images may become disordered and may no longer be in a sequential order (e.g., consistent with the original scan order). In order to put the short-axis MR slice images back into a sequential order (e.g., consistent with their original scan order), the bull's eye diagram generating apparatus may determine a position 502 where a short-axis MR slice image (e.g., a corresponding image plane of the short-axis MR slice image) intersects with a long-axis MR slice image of the heart, and use the distance from the position 502 to the landmark point 506 to determine the sequential order of the short-axis MR slice images.

[0027] The bull's eye image generation device may select a long-axis MR slice image 504 (e.g., an intermediate slice image) from a plurality of long-axis MR slice images obtained by the bull's eye image generation device, and determine a corresponding image plane based on metadata (e.g., header information such as DICOM header information), the metadata being associated with the short-axis and / or long-axis MR slice images, and the corresponding image plane corresponding to the short-axis MR slice image and the long-axis MR slice image 504. The metadata may be obtained by the bull's eye image generation device, for example, together with the short-axis and / or long-axis MR slice images, or separately from the short-axis and / or long-axis MR slice images. In response to obtaining the metadata, the bull's eye image generation device may extract information from the metadata, the information being related to the slice image / image orientation, the slice image / image position, the slice image / image pixel resolution, etc. The bull's eye image generation device may then use the extracted information to determine corresponding plane equations for the short-axis MR slice image and the long-axis MR slice image. Each of these equations may include a norm vector perpendicular to the corresponding slice image / image plane and a sampling point on the slice image / image plane (e.g., the norm vector and the sampling point may define the slice image / image plane). It can be seen that, using the corresponding norm vectors and sampling points associated with the short-axis MR slice image and the long-axis MR slice image 504, the bull's eye diagram generation device can determine the projection position 502 of the short-axis MR slice image on the long-axis slice image 504 (for example, the projection position can correspond to the intersection line of the short-axis MR slice image and the long-axis MR slice image 504). The bull's eye diagram generation device can then determine the corresponding distances of these projection positions 502 from the landmark point 506, and arrange (for example, sort) the short-axis MR slice images based on these distances. For example, given the short-axis MR slice images S1, S4, S3 and S2, the bull's eye diagram generation device can determine based on the metadata information associated with the short-axis slice image and / or the long-axis slice image 504 that S1, S4, S3 and S2 can be projected onto the long-axis slice image 504 at positions L1, L4, L3 and L2, respectively, and the distances of these positions from the landmark point 506 are D1, D4, D3 and D2, respectively. The bull's eye diagram generating device may also determine that the values ​​of the distances from the landmark point 506 are in descending order as D1, D2, D3, and D4. It can be seen from this that the bull's eye diagram generating device may determine that the sequential order of the short-axis MR slice images (e.g., consistent with their original scanning order) should be S1, S2, S3, and S4, wherein S1 may correspond to a slice image closer to the base of the myocardium (e.g., a base slice image), and S4 may represent a slice image closer to the apex of the myocardium (e.g., an apex slice image).

[0028] The bull's eye diagram generation device can also determine the direction (e.g., orientation) of the CMRI scan based on the projection position 502 of the short-axis MR slice image on the long-axis MR slice image 504. In an embodiment, if the first short-axis MR slice image has a projection position on the long-axis MR slice image 504 that is a greater distance from the landmark point 506 than the last short-axis MR slice image, the bull's eye diagram generation device can determine that the scan direction is from the base of the heart to the apex of the heart. In an embodiment, the bull's eye diagram generation device can determine the scan direction based on the average distance. For example, if the average distance of the projection position of slice images 1-5 from the landmark point 506 is greater than the average distance of the projection position of slice images 6-10 from the landmark point 506, the bull's eye diagram generation device can decide that the scan is from the base of the heart to the apex of the heart, or vice versa.

[0029] The bull's eye image generation device can determine the landmark point 506 and / or one or more other long-axis landmark points (e.g., a point indicating the level of the mitral valve annulus) based on the long-axis MR slice image obtained from the CMRI scan. The landmark point 506 can be, for example, the apex of the myocardium. The bull's eye image generation device can be configured to use the landmark detection neural network described herein or use a separate neural network to determine the landmark point 506 (e.g., the apex) and / or other long-axis landmark points to specifically identify the long-axis landmark points, the separate neural network having a similar structure to the landmark detection neural network but having been trained differently (e.g., using a different training data set and / or gold standard). For example, the bull's eye image generation device can be configured to determine the long-axis landmark points using a convolutional neural network, such as a deep convolutional neural network, including one or more convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Using these layers, the convolutional neural network can extract visual features from the long-axis slice image or image and identify areas or pixels that match the features of one or more long-axis landmark points that the neural network has learned from training.

[0030] Figure 6 Example operations that may be performed by a bull's eye image generation device described herein are illustrated. Operations may begin at 602, and the bull's eye image generation device may obtain a short-axis MR slice image and a long-axis MR slice image at 604 based on a CMRI scan performed on a patient. At 606, the bull's eye image generation device may segment (e.g., using a segmentation neural network described herein) an image included in the short-axis MR slice image to identify the myocardium, the left ventricle, and / or the right ventricle in the image. At 608, the bull's eye image generation device may determine (e.g., using a landmark detection neural network) a first landmark point, such as a first landmark point, based on the short-axis MR slice image and / or the segmentation. Figure 4 The bull's eye diagram generating device can also determine based on the long-axis MR slice image such as Figure 5The second landmark point of the apex 506 shown. At 610, the bull's eye map generation device can determine the corresponding projection positions of the short-axis MR slice images on a long-axis MR slice image based on, for example, metadata information associated with the CMRI scan. As described herein, these projection positions can be sequentially away from the identified landmark points (e.g., the apex of the myocardium) on the long-axis slice images. Therefore, the bull's eye map generation device can sequentially arrange the short-axis MR slice images based on the corresponding distances of the projection positions on the long-axis slice images from the identified landmark points at 612. At 614, the bull's eye map generation device can divide the myocardium depicted in the short-axis MR slice image into segments based on the first landmark point identified at 608, for example, Figure 4 Then, before the bull's eye map generation device ends the operation at 618 , the bull's eye map generation device may generate a bull's eye map (eg, a sixteen-segment or seventeen-segment bull's eye map) based on the myocardial segments and the sequentially ordered short-axis MR slice images.

[0031] The example operations have been depicted and described herein in a particular order. However, it should be noted that these operations may be performed in a different order, simultaneously, and / or with other operations not presented or described herein. Further, it should be noted that not all operations that a bull's-eye diagram generating device can perform are depicted and described herein, and not all illustrated operations need to be performed by a bull's-eye diagram generating device.

[0032] The neural networks described herein (e.g., segmentation neural networks, landmark detection neural networks, etc.) can be trained to optimize their parameters (e.g., weights associated with the layers of the neural network) for performing the various recognition, prediction, or estimation tasks described herein. The training can be performed using multiple images of the human heart and / or a corresponding loss function designed to guide the neural network through a learning and / or optimization process. Figure 7An example process 700 for training a neural network described herein (e.g., a segmentation neural network and / or a landmark detection neural network described herein) is illustrated. Process 700 may begin at 702, and at 704, the neural network may initialize its operating parameters, such as weights associated with one or more filters or kernels of the neural network. The parameters may be initialized, for example, based on samples of one or more probability distributions or parameter values ​​from another neural network having a similar architecture. At 706, the neural network may receive a training image of a human heart, process the image through various layers of the neural network, and make a prediction of a target result (e.g., a segmentation mask of the myocardium, LV, or RV, a classification of a landmark point, etc.) using the currently assigned parameters. At 708, the neural network may determine an adjustment to be made to the currently assigned parameters based on a loss function and a gradient descent (e.g., stochastic gradient descent) associated with the loss function. For example, the loss function may be implemented based on a mean squared error (MSE), a Dice ratio, a cross entropy, etc. between the prediction and a gold standard associated with the prediction (e.g., an annotation of the myocardium or a landmark point). At 710, the neural network may adjust the currently assigned parameters, for example, via a back-propagation process. At 712, the neural network can determine whether one or more training termination criteria are met. For example, if the neural network has completed a predetermined number of training iterations, if the difference between the predicted value and the gold standard value is below a predetermined threshold, or if the change in the value of the loss function between two training iterations is below a predetermined threshold, the neural network can determine that the training termination criteria are met. If it is determined at 712 that the training termination criteria are not met, the neural network can return to 706. If it is determined at 712 that the training termination criteria are met, the neural network can end the training process 700 at 714.

[0033] The bull's-eye diagram generating apparatus 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 8 800 is a block diagram illustrating an example bull's eye diagram generating device 800 described herein. As shown, the bull's eye diagram generating device 800 may include a processor 802, which may be a central processing unit (CPU), a graphics processing unit (GPU), a microcontroller, a reduced instruction set computer (RISC) processor, an application specific integrated circuit (ASIC), an application specific instruction set processor (ASIP), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), or any other circuit or processor capable of performing the functions described herein. The bull's eye diagram generating device 800 may also include a communication circuit 804, a memory 806, a mass storage device 808, an input device 810, and / or a communication link 812 (e.g., a communication bus), Figure 8One or more components shown can exchange information through the communication link. The communication circuit 804 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, wireless data networks (e.g., Wi-Fi, 3G, 4G / LTE, or 5G networks). The memory 806 may include a storage medium configured to store machine-readable instructions, which, when implemented, causes the processor 802 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.). The mass storage device 808 may include one or more disks, such as one or more built-in hard disks, one or more removable disks, one or more magneto-optical disks, one or more CD-ROM or DVD-ROM disks, etc., on which instructions and / or data may be stored to facilitate the operation of the processor 802. The input device 810 may include a keyboard, a mouse, a voice control input device, a touch-sensitive input device (eg, a touch screen), etc., for receiving user input of the bull's-eye diagram generating apparatus 800 .

[0034] It should be noted that the bull's-eye diagram generation apparatus 800 can operate as a standalone device or can be connected (e.g., networked or clustered) with other computing devices to perform the functions described herein. Figure 8 Only one example of each component is shown in the figure, and those skilled in the art will understand that the bull's eye diagram generation device 800 may include multiple instances of one or more components shown in the figure. In addition, although the examples are described herein with reference to various types of neural networks, various types of layers, and / or various tasks performed by certain types of neural networks or layers, these references are made for illustrative purposes only and are not intended to limit the scope of the present disclosure. And although the present disclosure has been described according to certain embodiments and generally associated methods, changes and transformations of the embodiments and methods will be obvious to those skilled in the art. Therefore, the above description of the exemplary embodiments does not limit the present disclosure. Other changes, substitutions and modifications are also possible without departing from the spirit and scope of the present disclosure. In addition, unless otherwise specifically stated, discussions using terms such as "analyze", "determine", "enable", "identify", "modify", etc. refer to the actions and processes of a computer system or similar electronic computing device, which manipulate and transform data represented as physical (e.g., electronic) quantities within the registers and memories of the computer system into other data represented as physical quantities within the computer system memory or other such information storage, transmission or display device.

[0035] It should be understood that the above description is intended to be illustrative, rather than restrictive. After reading and understanding the above description, many other embodiments will be apparent to those skilled in the art. Therefore, the scope of the present disclosure should be determined with reference to the appended claims and the full scope of equivalents to which such claims are assigned.

Claims

1. A device configured to generate a bull's eye chart, comprising: One or more processors configured to: obtaining a plurality of first MR slice images based on a first CMRI scan of the human heart taken along a short axis of the human heart; obtaining a second MR slice image based on a second CMRI scan of the human heart taken along the long axis of the human heart; Determine one or more first landmark points associated with the human heart based on the first MR slice image; wherein the one or more first landmark points indicate where a left ventricle and a right ventricle of the human heart intersect, or the one or more first landmark points further indicate a center of the left ventricle; determining a second landmark point associated with the human heart based on the second MR slice image; wherein the second landmark point indicates an apex of the myocardium of the human heart; determining a corresponding projection position of the first MR slice image on the second MR slice image; sequentially arranging the first MR slice images based on respective distances from the projection positions of the first MR slice images on the second MR slice images to the second landmark points; determining a plurality of myocardial segments based on the sequentially arranged first MR slice images and the one or more first landmark points; and The bull's eye plot is generated based on the plurality of myocardial segments.

2. The device according to claim 1, wherein The one or more processors are configured to determine a first portion of the myocardium and a second portion of the myocardium based on the intersection of the left ventricle and the right ventricle, and to divide the first portion of the myocardium into a first group of one or more equal segments based on the center of the left ventricle and the intersection of the left ventricle and the right ventricle, and to divide the second portion of the myocardium into a second group of one or more equal segments.

3. The device according to claim 1, wherein: The one or more processors are configured to segment at least one of the first MR slice images to identify the left ventricle, the right ventricle, and the myocardium in the at least one of the first MR slice images, and determine the one or more first landmark points based on the segmentation.

4. The device according to claim 3, wherein: The one or more processors are configured to segment the at least one of the first MR slice images using an artificial neural network to identify the left ventricle, the right ventricle, and the myocardium.

5. The device according to claim 1, wherein: The one or more processors are configured to arrange the first MR slice images from a basal slice image to an apical slice image based on the corresponding projection positions of the first MR slice images on the second MR slice image, the basal slice image corresponding to the position on the second MR slice image having the longest distance to the apex, and the apical slice image corresponding to the position on the second MR slice image having the shortest distance to the apex.

6. The device according to claim 1, wherein: The one or more processors are further configured to determine the first CMRI scanning order based on the corresponding projection positions of the first MR slice images on the second MR slice images.

7. The device according to claim 1, wherein: The one or more processors are configured to obtain metadata associated with the first MR slice image and the second MR slice image, and determine the corresponding projection position of the first MR slice image on the second MR slice image based on the metadata.

8. A method for generating a bull's eye chart, comprising: obtaining a plurality of first MR slice images based on a first CMRI scan of the human heart taken along a short axis of the human heart; obtaining a second MR slice image based on a second CMRI scan of the human heart taken along the long axis of the human heart; Determine one or more first landmark points associated with the human heart based on the first MR slice image; wherein the one or more first landmark points indicate where a left ventricle and a right ventricle of the human heart intersect, or the one or more first landmark points further indicate a center of the left ventricle; determining a second landmark point associated with the human heart based on the second MR slice image; wherein the second landmark point indicates an apex of the myocardium of the human heart; determining a corresponding projection position of the first MR slice image on the second MR slice image; sequentially arranging the first MR slice images based on respective distances from the projection positions of the first MR slice images on the second MR slice images to the second landmark points; determining a plurality of myocardial segments based on the sequentially arranged first MR slice images and the one or more first landmark points; and The bull's eye plot is generated based on the plurality of myocardial segments.

Images