Methods for motion artifact detection

Abstract

This disclosure relates to a medical imaging method for motion artifact detection. The method includes: generating a motion-damaged image with motion artifacts such as those caused by a first initial motion pattern using (201-203) k-space acquisition properties, such that the motion artifacts are defined as a function of a feature matrix and the motion-damaged image; initializing at least one feature map of a convolutional neural network (CNN) using the values ​​of a convolutional matrix (205); training the initialized CNN (207) to obtain motion artifacts caused by a second training motion pattern in a training image; and using the trained CNN to obtain (209) motion artifacts in an input image.

Description

Technical Field

[0001] This invention relates to scanning imaging systems, and more particularly to a method for detecting motion artifacts in input images. Background Technology

[0002] During a magnetic resonance imaging (MRI) scan, radio frequency (RF) pulses generated by a transmitter antenna cause disturbances to the local magnetic field, and RF signals emitted by nuclear spins are detected by a receiver antenna. These RF signals are used to construct MR images. During longer scans, the scanned object may exhibit internal or external motion, which corrupts the data and results in MR images with blurriness or artifacts. Many methods have been proposed to mitigate or correct MR motion artifacts. These methods, however, operate in image space and attempt to identify specific features of motion artifacts based on the affected image. Summary of the Invention

[0003] Various embodiments provide a method, medical analysis system, and computer program product for detecting motion artifacts in input images, as described in the subject matter of the independent claims. Advantageous embodiments are described in the dependent claims.

[0004] Motion artifacts are one of the most common causes of image degradation in MRI. Deep learning (DL) techniques can be applied to motion correction of MR images, for example, by applying a DL network in image space after conventional MR image reconstruction. However, these techniques may discard explicit information about the k-space sampling pattern and its timing. Some, but not all, of this information is implicitly learned during the training of the DL network's convolutional kernels after initialization with noise values. This can require considerable effort and a large amount of training data. This topic also explores ways to improve motion correction by using explicit knowledge of the sampling pattern and its timing to design DL networks and initialize their kernels so that they are pre-trained to estimate common motion artifacts, such as translational motion artifacts. These kernels can be computed with low effort directly from sequence parameters and predefined motion paths using algorithms. This initialization can significantly reduce the effort required for training, which is then performed to address other forms of motion.

[0005] In one aspect, the present invention relates to a medical imaging method for motion artifact detection, and more particularly to a method for detecting motion artifacts in an input image. The method includes: generating a motion-damaged image with motion artifacts such as those caused by a first initial motion pattern using k-space acquisition properties, such that the motion artifacts are a function of a feature matrix and the motion-damaged image; initializing at least one feature map of a convolutional neural network (CNN) using the values ​​of the feature matrix; training the initialized CNN using training images, the training images being subjected to motion damage of a second pattern (type) to obtain motion artifacts; and using the trained CNN to obtain motion artifacts in the input image. The motion of the first initial setting type and the second motion pattern are completely approximated by multiple displacements. The approximation accuracy of the initial motion pattern (the motion of the first initial setting type) and the motion pattern of the image to be corrected for the artifacts (the motion (pattern) of the second training type) depends on a predetermined level of residual artifacts associated with a preset desired image quality.

[0006] This topic enables the network to learn the pure properties of the artifacts rather than also learning anatomical properties. The CNN is initialized with image features of MR motion artifacts for motion correction, thus reducing training effort. The obtained motion artifacts can be used, for example, to correct motion artifacts in the input image. This can increase the accuracy of artifact detection and therefore the accuracy of MR motion correction. This topic enables automatic correction of motion artifacts in magnetic resonance images.

[0007] This topic can prevent the loss of artifact information by using k-space acquisition properties. In particular, this topic takes into account the fact that any motion caused by continuous acquisition patterns in k-space is transformed into motion-degraded temporal coherence in k-space, which is much more difficult to “learn” through convolutional networks in the image domain, and explicit knowledge about temporal coherence is lost during Fourier reconstruction.

[0008] For example, the trained CNN can be accessible by a remote computer system (e.g., the trained CNN can be stored in the computer system), wherein the computer system is configured to automatically detect motion artifacts in the MR images using the trained CNN after receiving them. This can be advantageous because it enables centralized and consistent artifact detection using the trained CNN, for example, across different MRI systems.

[0009] The k-space acquisition properties can be determined according to an MR imaging protocol. An imaging protocol or protocol refers to a set of technical settings or a set of acquisition parameters for an imaging modality to produce images (e.g., MR images) required for an examination. For example, acquisition parameters may indicate the type of pulse sequence. For example, the method includes providing a given MR imaging protocol, wherein the k-space acquisition properties can be determined using the given MR imaging protocol. The k-space characteristics can be determined by the k-space sampling mode and its timing. A set of object displacements dx(t) can be selected such that they reflect typical real-world motion (e.g., of the first type), for example, rapid displacement to a new position at a constant velocity in any direction, after an initial period of stillness, oscillation, or more complex motion. The maximum amplitude of dx(t) can be much smaller than the field of view (FOV) of an image in a clinical imaging setting. All selections of dx(t) can conform to the laws of mechanics, for example, being a continuous and quadratically differentiable function of time. This can be advantageous because it can also increase the accuracy of artifact detection by using a consistent data acquisition setting while allowing for different artifacts caused by motion. Moreover, training can enable the learning of artifacts of rotational and non-rigid motion.

[0010] According to one embodiment, the initialization includes: for each feature map of the CNN, determining a feature matrix with different first initial motion modes, and using the values ​​of the feature matrix to initialize the feature map.

[0011] For example, each assumed motion type yields an individual feature matrix. This embodiment enables the use of individual motions to initialize the individual kernels (or feature maps) of the parallel kernels of the CNN. Preferably, large motions can be assigned to low-resolution kernels, and conversely, small motions to high-resolution kernels. For example, the first initial motion pattern is determined based on the resolution of the kernel to be initialized. If the kernel to be initialized has a resolution higher than a predefined resolution threshold, the first initial motion pattern may have a motion amplitude or shift lower than a predefined motion amplitude threshold. If the kernel to be initialized has a resolution lower than a predefined resolution threshold, the first initial motion pattern may have a motion amplitude or shift higher than a predefined motion amplitude threshold. In another example, for each resolution of the CNN kernel, a corresponding first motion type can be defined and used to define the motion artifacts. This enables accurate motion artifact detection because the CNN is initialized accurately on an individual basis.

[0012] According to one embodiment, the motion artifacts are defined in k-space as the product of the feature matrix and the motion-damaged image, wherein the feature map is a multiplication layer of a CNN, and the CNN is configured to receive a k-space representation of the image as input. This embodiment enables the CNN to operate entirely in k-space (the CNN has Fourier Transform (FT) layers) and is directly fed with measurement data corresponding to the FT of the measured image. The CNN in this embodiment may include multiplication layers instead of convolutional kernels.

[0013] According to one embodiment, the motion artifacts are defined in image space as a convolution of the feature map and the motion-damaged image, wherein the feature map is the kernel of a CNN, and the CNN is configured to receive the image as input. This embodiment enables the implementation of a CNN to operate entirely in image space and be directly fed with the measurement image. The CNN in this embodiment may include convolutional kernels.

[0014] In another embodiment, the CNN may include FT layers and FT representing k-space and image space, respectively. -1 Each layer of the CNN initializes its feature map using a feature matrix obtained in the corresponding space. The CNN can then enable switching between image space and k-space in subsequent layers.

[0015] According to one embodiment, the first initial motion pattern is a translational motion characterized by a corresponding translational displacement. This embodiment may be advantageous because the translational motion can be a good first approximation of all current motions during MR scanning. Therefore, the effort and amount of training for this type of motion can be reduced by this subject matter. In particular, this subject matter can reduce or save the effort involved in the following two situations: 1) The computation of training data typically involves simulating motion artifacts over a large number of individual images, involving per-image FT, k-space data manipulation, and inverse FT. 2) Actual training may also require considerable effort to progressively “teach” the convolutional kernel using a random subset of training images.

[0016] According to one embodiment, the second training motion pattern is a combination of first initial motion patterns, wherein each first initial motion pattern is characterized by a corresponding translational displacement.

[0017] According to one embodiment, the motion artifacts in the input image are caused by motion, which is a combination of a first initial motion pattern and a second training motion pattern; or the second training motion pattern.

[0018] According to one embodiment, the first type of exercise may be different from or the same as the second type of exercise. The second training exercise mode may be a combination of different first initial exercise modes.

[0019] According to one embodiment, the k-space acquisition properties include the sampling pattern and / or sampling timing of the k-space. This subject matter can, for example, use explicit knowledge of the sampling pattern and its timing to design and initialize a CNN, such that it can be set up and pre-trained to estimate translational motion artifacts in an image. The CNN can also be trained to handle rotations and more complex motions. This contrasts with another approach that operates entirely in image space, resulting in the loss of at least some information about the k-space sampling pattern and its timing. There are structures and information that may not be utilized by this other approach, for example, because the probability that two sampling points in k-space represent the same motion state decreases continuously with their distance in the sampling time. Even if implicit knowledge of the pattern and its timing is introduced into the network during training in other methods, it may be insufficient or ineffective because it is indirect and therefore inefficient, and it may not utilize all available information.

[0020] According to one embodiment, the motion artifact A of the motion-damaged image M is defined as the convolution of the motion-damaged image M and the feature matrix K, wherein the feature matrix K is composed of... Define, where, δ is the δ function, dx(t) is the motion function, FT represents the Fourier transform, and k is the k-space location of the motion-damaged image M.

[0021] According to one embodiment, the motion artifact A of the motion-damaged image M in k-space is defined as the product of the motion-damaged image M and the feature matrix in k-space, wherein the feature matrix is ​​composed of... Define, where, δ is the δ function, dx(t) is the motion function, FT represents the Fourier transform, and k is the k-space location of the motion-damaged image M.

[0022] This embodiment of the CNN may include multiplication layers instead of convolutional kernels, and the convolutional kernels may be represented by feature matrices. Initialization, such as that derived from the properties of k-space, including k-space sampling patterns and their timing and specific motion.

[0023] According to one embodiment, the k-space representation is a two-dimensional or three-dimensional representation.

[0024] According to one embodiment, the method further includes receiving the input image, particularly a medical image, from a magnetic resonance imaging (MRI) system, the method being performed during operation of the MRI system. This enables online or real-time artifact correction methods.

[0025] Another aspect of the invention relates to the training of a convolutional neural network, including the separate initial settings of its kernels (feature maps) and the training of the CNN. Compared to conventional training, the training of the CNN according to the invention requires less effort and / or time. Subsequently, the trained CNN can be used in motion correction of magnetic resonance images.

[0026] In another aspect, the present invention relates to a computer program product comprising machine-executable instructions for execution by a processor, wherein execution of the machine-executable instructions causes the processor to perform the method according to any one of the preceding claims.

[0027] In another aspect, the present invention relates to a medical analysis system, the medical imaging system including at least one processor; and at least one memory including computer program code; the at least one memory and the computer program code being configured to use the at least one processor to cause the medical imaging system to perform at least a portion of the method according to any of the foregoing embodiments.

[0028] The medical analysis system is configured to connect to multiple MRI systems and receive the input images from the MRI systems.

[0029] In another aspect, the present invention relates to an MRI system including a medical analysis system. The MRI system is configured to acquire image data and reconstruct an initial image from the image data, the initial image being processed by the medical analysis system to perform at least a portion of the method according to any one of the preceding claims.

[0030] It should be understood that one or more of the foregoing embodiments of the present invention may be combined, as long as the combined embodiments are not mutually exclusive. Attached Figure Description

[0031] Preferred embodiments of the invention will be described below by way of example only and with reference to the accompanying drawings, wherein:

[0032] Figure 1 This is a schematic diagram of a medical analysis system.

[0033] Figure 2 This is a flowchart of a medical imaging method for motion artifact detection;

[0034] Figure 3 This is a flowchart illustrating a method for determining the convolution matrix, based on an example from this topic.

[0035] Figure 4 This is a flowchart illustrating a method for verifying the convolution matrix determination approach, based on an example from this topic.

[0036] Figure 5Images illustrating methods for detecting artifacts according to this topic are depicted.

[0037] Figure 6 Images illustrating methods for detecting artifacts according to this topic are depicted.

[0038] Figure 7 A cross-sectional and functional view of the MRI system is shown.

[0039] List of reference numerals

[0040] 100 Medical Systems

[0041] 101 Scan Imaging System

[0042] 103 processor

[0043] 107 Memory

[0044] 108 power supply

[0045] 109 bus

[0046] 111 Control System

[0047] 121 Software

[0048] 125 monitor

[0049] 129 User Interface

[0050] 150 AI components

[0051] 201-413 Method and Steps

[0052] 700 Magnetic Resonance Imaging System

[0053] 704 Magnet

[0054] 706 Magnet Chamber

[0055] 708 Imaging Area

[0056] 710 Magnetic Gradient Coil

[0057] 712 Magnetic Gradient Coil Power Supply

[0058] 714 RF Coil

[0059] 715 RF Amplifier

[0060] 718 objects Detailed Implementation

[0061] In the following figures, similarly numbered elements are either similar elements or perform equivalent functions. If the functions are equivalent, elements already discussed will not need to be discussed in the following figures.

[0062] Various structures, systems, and devices are schematically depicted in the accompanying drawings for illustrative purposes only and so as not to obscure the invention with details well known to those skilled in the art. However, the drawings are included to describe and explain illustrative examples of the disclosed subject matter.

[0063] Figure 1 This is a schematic diagram of a medical analysis system 100. The medical analysis system 100 includes a control system 111 configured to connect to a scanning imaging system (or acquisition unit) 101. The control system 111 includes a processor 103 and a memory 107, each capable of communicating with one or more components of the medical system 100. For example, components of the control system 111 are coupled to a bidirectional system bus 109.

[0064] It will be appreciated that the methods described herein are at least partially non-interactive and automated through computerized systems. For example, these methods can also be implemented in software 121 (including firmware), hardware, or a combination thereof. In exemplary embodiments, the methods described herein are implemented as software as an executable program and executed by a dedicated or general-purpose digital computer, such as a personal computer, workstation, minicomputer, or mainframe computer.

[0065] Processor 103 is a hardware device for running software specifically stored in memory 107. Processor 103 can be any custom or commercially available processor, central processing unit (CPU), auxiliary processor among several processors associated with control system 111, semiconductor-based microprocessor (in the form of a microchip or chipset), microprocessor, or any device typically used to run software instructions. Processor 103 can control the operation of scanning imaging system 101.

[0066] Memory 107 may include any or a combination of volatile memory elements (e.g., random access memory (RAM, such as DRAM, SRAM, SDRAM, etc.)) and non-volatile memory elements (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM)). It should be noted that memory 107 may have a distributed architecture, wherein various components are located remotely from each other but are accessible by processor 103. Memory 107 may store instructions or data relating to at least one other constituent element of the medical system 100.

[0067] The control system 111 may also include a display device 125, which displays characters and images, for example, on the user interface 129. The display device 125 may be a touch screen display device.

[0068] The medical analysis system 100 may also include a power supply 108 for powering the medical analysis system 100. The power supply 108 may be, for example, a battery or an external power source, such as power supplied from a standard AC outlet.

[0069] The scanning imaging system 101 may include at least one of an MRI, CT, and PET-CT imager. The control system 111 and the scanning imaging system 101 may or may not be integral parts. In other words, the control system 111 may or may not be external to the scanning imaging system 101.

[0070] The scanning imaging system 101 includes components that can be controlled by the processor 103 to configure the scanning imaging system 101 to provide image data to the control system 111. The configuration of the scanning imaging system 101 enables its operation. The operation of the scanning imaging system 101 can be, for example, automatic. Figure 7 An example of a component of the scanning imaging system 101 in an MRI system is shown.

[0071] The connection between the control system 111 and the scanning imaging system 101 can include, for example, a BUS Ethernet connection, a WAN connection, an Internet connection, etc.

[0072] In one example, the scanning imaging system 101 can be configured to provide output data, such as images, in response to specified measurement results. The control system 111 can be configured to receive data from the scanning imaging system 101, such as MR image data. For example, the processor 103 can be adapted to receive information from the scanning imaging system 101 in a compatible digital form (automatically or upon request), such information being displayed on the display device 125. Such information may include operating parameters, alarm notifications, and other information relating to the purpose, operation, and function of the scanning imaging system 101.

[0073] The medical analysis system 100 can be configured to communicate with other scanning imaging systems 131 and / or databases 133 via a network 130. The network 130 includes, for example, a wireless local area network (WLAN) connection, a WAN (wide area network) connection, a LAN (local area network) connection, or a combination thereof. The database 133 may include information related to patients, scanning imaging systems, anatomical structures, scanning geometry, scanning parameters, scans, etc. The database 133 may include, for example, an EMR database containing patient EMR data, a radiology information system database, a medical image database, a PACS, a hospital information system database, and / or other databases comparing data that can be used to plan scanning geometry. The database 133 may include, for example, training images. Additionally or alternatively, training images may be stored in the local storage of the control system 111 (e.g., disk storage or memory).

[0074] The memory 107 may also include an artificial intelligence (AI) component 150. The AI ​​component 150 may or may not be part of the software component 121. The AI ​​component 150 may be configured to train a CNN based on the subject matter and provide the trained CNN for further use. For example, if the control system 111 is not part of the scanning imaging system 101, the trained CNN may be provided to the scanning imaging system 101 so that it can be used at the scanning imaging system 101 to detect artifacts in the image reconstructed by the scanning imaging system 101.

[0075] Figure 2 This is a flowchart of a medical imaging method for motion artifact detection. In step 201, a motion-damaged image M can be obtained. The motion-damaged image can be a modeled or simulated image; for example, this can be performed starting from an ideal image without motion artifacts. The motion-damaged image can be generated based on a given k-space acquisition property. The k-space acquisition property can, for example, include a sampling pattern x and its timing t. One or more shift functions dx(t) describe a first initial motion pattern. The first initial motion pattern can be a translational displacement function dx(t), where t can include all time during which the k-space is sampled. The motion-damaged image has known motion artifacts caused by the first initial motion pattern.

[0076] Motion artifact A can be defined or modeled in step 203 as a function of the feature matrix and the motion-damaged image M. Motion artifact A can be defined in k-space and / or image space. In one example, the motion artifact A of the motion-damaged image M can be modeled in image space as the convolution of the motion-damaged image M with the feature matrix. In this case, the feature matrix is ​​the convolution matrix. In another example, the motion artifact A of the motion-damaged image M can be modeled in k-space as the product of the motion-damaged image M with the feature matrix. In this case, the feature matrix is ​​the multiplication matrix.

[0077] In one example, steps 201 and 203 can be repeated for each feature map of the CNN using a different first initial motion pattern. This allows each feature map of the CNN to be associated with a corresponding feature matrix. The differences between the first initial motion patterns of these matrices can be defined based on, for example, the resolution of the feature maps. For example, large motions can be assigned to low-resolution feature maps, and conversely, small motions can be assigned to high-resolution feature maps.

[0078] At least one feature map of the CNN can be initialized in step 205 using the values ​​of the feature matrix. In one example, each filter kernel of the CNN can be initialized using the values ​​of the feature matrix. This allows for improved training of the CNN and thus improved accuracy in artifact detection. In another example, each filter kernel of the first layer of the CNN can be initialized using the values ​​of the feature matrix. This also improves the training accuracy of the CNN and the accuracy of artifact detection. In yet another example, for each filter kernel of the CNN, steps 201-205 can be repeated for different first-type motions, resulting in each filter kernel being initialized with the corresponding feature matrix. For example, multiple motion-damaged images can be obtained using different hypothesis shift functions dx(t), and the corresponding feature matrices can be determined. This also improves the training accuracy of the CNN and the accuracy of artifact detection. In one example, the feature maps of the CNN can have corresponding predefined sizes. In another example, the size of each feature map of the CNN can be defined using a first initial motion pattern for initializing the feature maps. This allows the feature map to be no larger than the range of a first initial motion pattern in a first direction (e.g., the x-direction), and in a second direction (e.g., the y-direction), the size of the feature map is controlled by k-space timing and the acquisition mode. This is, for example, in... Figure 5 and Figure 6 The image representing "kernel K" is shown, where the white pixels of the image can be used to determine the size of the feature map to be initialized by "kernel K". For example, equation (8) calculates K at full resolution for a given motion and then scales it down to the relevant support such that it corresponds to the size of the feature map. The scaled-down version can be used to initialize the feature map of the CNN.

[0079] The initialized CNN can be trained in step 207 to obtain motion artifacts caused by the second trained motion pattern in the training images. Each training image has a motion artifact caused by the second trained motion pattern. In one example, the second trained motion pattern can be a combination of multiple first initial motion patterns, where each first initial motion pattern is represented by a corresponding translational displacement. This can improve the accuracy of CNN training and artifact detection because the initialization has already taken into account at least a portion of the motion in the training images. Training can converge rapidly by using this initialization method.

[0080] Using a trained CNN, motion artifacts can be obtained from the input image in step 209.

[0081] Obtaining motion artifacts can, for example, involve using a CNN to determine whether an input image has motion artifacts, and if so, using values ​​provided by a trained CNN to obtain the motion artifacts. For instance, a trained CNN can provide multiple pixel values ​​as output, where each pixel value indicates the content of the motion artifact at that pixel location, which can be zero or absent. In other words, the CNN can take a measured motion-damaged image as input and estimate the pure motion artifact as output. The pure artifact can be subtracted from the measured image to produce a corrected image. Motion artifact detection using a trained CNN is likely to be accurate because it is initialized with values ​​reflecting the real motion that occurs anomalously during MR imaging. This can be particularly advantageous if the motion artifacts in the input image are caused by a combination of a first initial motion pattern and a second trained motion pattern; or by motion as the second trained motion pattern.

[0082] Figure 3 This is a flowchart illustrating a method for determining a convolution matrix, based on an example from this topic.

[0083] For example, suppose an ideal image I and its k-space data S = FT(I) undergo motion characterized by a translation function dx(t), where t includes all time when the k-space is sampled. The motion-damaged image M can be determined or defined in step 301 as follows: Where E is the exponential phase term.

[0084] The motion-damaged image in equation (1) can be defined in step 303 as a function of kernel D that generates image M from ideal image I. This can be performed using equations (2) to (3) as follows.

[0085]

[0086] If the inverse FT of E is referred to as D = FT -1 (E), and If it's a convolution, then M can be written as:

[0087]

[0088] D is the inverse Fourier transform of the exponential phase term E defined by the k-space sampling mode, its timing, and motion. If A is a pure artifact within the motion-damaged image M, and δ is a function of δ, then the pure artifact A can be defined in step 305 using image M as follows.

[0089]

[0090]

[0091] This led to

[0092] The convolution matrix can be determined in step 307 by first applying the Furrier transform to equation (5) using the convolution theorem. This yields equation (6).

[0093] FT(A)FT(D)=FT(M)FT(D-δ) (6)

[0094]

[0095] Furthermore, the inverse Fourier transform is applied to equation (6') using the convolution theorem. This yields equation (7).

[0096] in,

[0097] Equation (8) provides a model for the convolution matrix K. The convolution matrix K in image space is the kernel, which can be used to produce pure artifacts each time it is convolved with the measured image. Since the kernel D is the inverse FT of the exponential phase term E, which depends on the k-space sampling mode, its timing, and motion, the convolution matrix K may also depend on these properties individually.

[0098] The convolution matrix K in equation (8) can be determined individually for each combination of motion and imaging protocol. For example, the motion type is defined by the value of dx(t), while the sampling mode and timing are defined by the set of acquisition parameters of the imaging protocol (e.g., by acquisition type).

[0099] Figure 4 It is used for checking or verifying. Figure 3 The flowchart of the method. Figure 4 An example of acquisition that includes motion, as incorporated in equation (8), can be used. For this purpose, an ideal phantom image I of the disk is used to simulate the motion-damaged image M. Furthermore, it is assumed that a selected specific motion dx(t) occurs during acquisition using a fast field echo sequence with typical fast factors, sampling patterns, and timing. For example, Figure 4The method may include the following steps for each set of motion and sequence parameters. In step 401, a time map T is calculated, which assigns sampling times to each point in the k-space based on the sequence type and parameters. In step 403, a phase map caused by motion dx(t) in the k-space is calculated using the exponent in equation (2). In step 405, a motion-damaged image M is calculated using equations (2) and (3). In step 407, a standard data artifact A1 is calculated by subtracting the ideal phantom image I from M. In step 409, a convolution matrix K for the initial setup of the CNN is calculated using equation (8). In step 411, a pure artifact A2 is calculated using the convolution matrix K and equation (7). In step 413, the difference between A2 and the standard data A1 is calculated to evaluate the accuracy of equations (7) and (8), and thus to evaluate Figure 3 The accuracy of the method. Figure 5 and Figure 6 It shows how to use Figure 4 The image obtained as described in the steps.

[0100] Figure 5 The data depicts the acquisition based on TFE sequences and segmented Cartesian k-space (resolution 256). 2 The results of the inspection method (TR 9ms, fast factor 4, echo interval 2ms) are shown. Assume that after acquiring 70% of the k-space data, there is a sudden shift of 5 pixels in the readout direction. Figure 5 As shown, the temporal plot visualizes the 4-fold segmentation in k-space. The phase plot shows the flat phase of 70% of the data and the linear phase in the readout direction, where the slope of the phase encoding direction is increased due to the displacement of the remaining data. The pure artifact A2 calculated using the convolution matrix K is almost identical to the artifact A1 calculated by subtracting the ideal image I from the measured image M, for example, by an MRI system 700. The convolution matrix K shows two rows of pixels with a displacement of 5 pixels, which approximates the prescribed motion.

[0101] Figure 6 This illustrates data for continuous motion over 10 pixels in the readout direction with a fast factor of 8 (TR 17ms, inter-echo time 2ms). The convolution matrix K is analogous to this motion via a 10-pixel-long horizontal line, now vertically spaced further due to the higher segmentation. An additional 5 pixels of motion in the phase-encoded direction are added.

[0102] exist Figure 5 and Figure 6 In the example, the convolution matrix K is primarily analogous to the motion path, which is replicated in the phase encoding direction in a manner that depends on the specific sampling mode. In all examples, the error in calculating artifacts using kernels is negligible, which proves the correctness of equations (7) and (8).

[0103] Figure 7The illustration shows a magnetic resonance imaging (MRI) system 700 as an example of a medical system 100. The MRI system 700 includes a magnet 704. The magnet 704 is a superconducting cylindrical magnet having a bore 706 therein. The use of different types of magnets is also possible; for example, both split cylindrical magnets and so-called open or sealed magnets are possible. A split cylindrical magnet is similar to a standard cylindrical magnet, except that the cryostat has been split into two segments to allow access to the isoplanar region of the magnet. Such a magnet can be used, for example, in conjunction with charged particle beam therapy. An open magnet has two magnet segments, one on top of the other, with a sufficiently large space between them to receive the object 718 to be imaged: the arrangement of the two segment regions is similar to the arrangement of Helmholtz coils. An assembly of superconducting coils is located inside the cryostat of the cylindrical magnet. Within the bore 706 of the cylindrical magnet 704, there is an imaging region or volume or anatomical structure 708, in which the magnetic field is sufficiently strong and homogeneous to perform MRI.

[0104] Within the bore 706 of the magnet, there is also an assembly of magnetic field gradient coils 710, which are used to spatially encode the magnetic spin of a target volume within the imaging volume 704 or examination volume 708 of the magnet during magnetic resonance data acquisition. The magnetic field gradient coils 710 are connected to a magnetic field gradient coil power supply 712. The magnetic field gradient coils 710 are intended to be representative. Typically, the magnetic field gradient coils 710 comprise three separate coil sets for encoding in three orthogonal spatial directions. The magnetic field gradient power supply supplies current to the magnetic field gradient coils. The current supplied to the magnetic field gradient coils 710 is time-controlled and can be either slanted or pulsed.

[0105] The MRI system 700 also includes an RF coil 714 located at the object 718 and adjacent to the examination volume 708 for generating RF excitation pulses. The RF coil 714 may include, for example, a set of surface coils or other specialized RF coils. The RF coil 714 can be used alternately for RF pulse transmission and for receiving magnetic resonance signals; for example, the RF coil 714 may be implemented as a transmit array coil including multiple RF transmit coils. The RF coil 714 is connected to one or more RF amplifiers 715.

[0106] The magnetic field gradient coil power supply 712 and RF amplifier 715 are connected to the hardware interface of the control system 111. The memory 107 of the control system 111 may, for example, include a control module. The control module contains computer-executable code that enables the processor 103 to control the operation and functions of the magnetic resonance imaging system 700. It also implements the basic operations of the magnetic resonance imaging system 700, such as the acquisition of magnetic resonance data.

[0107] As those skilled in the art will appreciate, various aspects of the present invention can be implemented as apparatus, method, or computer program product. Accordingly, various aspects of the present invention can take the form of a completely hardware embodiment, a completely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects (all of which may be referred to herein as "circuit," "module," or "system" in general). Furthermore, various aspects of the present invention can take the form of a computer program product implemented in one or more computer-readable media having computer-executable code implemented thereon.

[0108] Any combination of one or more computer-readable media can be used. The computer-readable media can be a computer-readable signal medium or a computer-readable storage medium. As used herein, "computer-readable storage medium" encompasses any tangible storage medium capable of storing instructions executable by a processor of a computing device. A computer-readable storage medium may be referred to as a computer-readable non-transitory storage medium. A computer-readable storage medium may also be referred to as a tangible computer-readable medium. In some embodiments, a computer-readable storage medium may also be capable of storing data accessible by a processor of a computing device. Examples of computer-readable storage media include, but are not limited to: floppy disks, magnetic hard disk drives, solid-state drives, flash memory, USB thumb drives, random access memory (RAM), read-only memory (ROM), optical discs, magneto-optical discs, and processor register files. Examples of optical discs include compact discs (CDs) and digital universal discs (DVDs), such as CD-ROMs, CD-RWs, CD-Rs, DVD-ROMs, DVD-RWs, or DVD-R discs. The term computer-readable storage medium also refers to various types of recording media accessible by a computer device via a network or communication link. For example, data can be retrieved on a modem, the Internet, or a local area network. Computer-executable code implemented on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, fiber optic cable, RF, or any suitable combination thereof.

[0109] Computer-readable signal media may include propagated data signals having computer-executable code implemented therein, for example, in baseband or as a carrier wave. Such propagated signals may take any variety of forms, including but not limited to electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that is not a computer-readable storage medium and is capable of conveying, propagating, or transmitting a program used by or in conjunction with an instruction execution system, apparatus, or device.

[0110] "Computer memory" or "memory" is an example of a computer-readable storage medium. Computer memory is any memory that can be directly accessed by a processor. "Computer storage device" or "storage device" is another example of a computer-readable storage medium. A computer storage device is any non-volatile computer-readable storage medium. In some embodiments, a computer storage device may also be computer memory, or vice versa.

[0111] As used herein, the term "processor" encompasses electronic components capable of executing programs or machine-executable instructions or computer-executable code. References to computing devices including "processor" should be interpreted as capable of containing more than one processor or processing core. A processor may, for example, be a multi-core processor. A processor may also refer to a collection of processors within a single computer system or distributed across multiple computer systems. The term computing device should also be interpreted as capable of referring to a collection or network of computing devices, each comprising one or more processors. Computer-executable code can be executed by multiple processors, which may be within the same computing device or even distributed across multiple computing devices.

[0112] Computer executable code may include machine-executable instructions or programs that instruct a processor to perform aspects of the present invention. Computer executable code for performing operations related to aspects of the present invention may be written in any combination of one or more programming languages ​​and compiled into machine-executable instructions, including object-oriented programming languages ​​such as Java, Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages. In some instances, the computer executable code may be in the form of a high-level language or in a pre-compiled form and used in conjunction with an interpreter that generates machine-executable instructions at runtime.

[0113] The computer-executable code may be executed entirely on the user's computer, partially on the user's computer (as a standalone software package), partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet provided by an Internet service provider).

[0114] Aspects of the invention are described with reference to flowchart illustrations, diagrams, and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that, when applicable, each block or portion of a flowchart illustration, diagram, and / or block diagram can be implemented by computer program instructions in the form of computer-executable code. It should also be understood that combinations of blocks from different flowchart illustrations, diagrams, and / or block diagrams can be combined without mutual exclusion. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus that produces the machine, such that the instructions, executable via the processor of the computer or other programmable data processing apparatus, create units for implementing the functions / actions specified in the flowchart illustrations and / or one or more block diagram blocks.

[0115] These computer program instructions may also be stored in a computer-readable medium that can instruct a computer, other programmable data processing apparatus or other device to operate in a particular manner, such that the instructions stored in the computer-readable medium produce an article of writing that includes instructions that implement the functions / actions specified in flowcharts and / or one or more block diagrams.

[0116] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device, thereby producing a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide a process for the function / action specified in the flowchart and / or one or more block diagram boxes.

[0117] As used herein, a "user interface" is an interface that allows a user or operator to interact with a computer or computer system. A "user interface" can also be referred to as a "human-machine interface device." A user interface can provide or receive information or data from an operator. A user interface enables input from an operator to be received by the computer and output from the computer to the user. In other words, the user interface allows an operator to control or manipulate the computer, and the interface allows the computer to indicate the effects of the operator's control or manipulation. The display of data or information on a monitor or graphical user interface is an example of providing information to an operator. The reception of data via a keyboard, mouse, trackball, touchpad, pointing stick, graphics tablet, joystick, game controller, webcam, headset, gearshift, steering wheel, pedals, wired gloves, dance mat, remote control, and accelerometer are all examples of user interface components that implement the reception of information or data from an operator.

[0118] As used herein, "hardware interface" encompasses the interface that enables a computer system's processor to interact with and / or control external computing devices and / or devices. A hardware interface can allow the processor to send control signals or instructions to external computing devices and / or devices. A hardware interface can also enable the processor to exchange data with external computing devices and / or devices. Examples of hardware interfaces include, but are not limited to: Universal Serial Bus (USB), IEEE 1394 port, parallel port, IEEE 1284 port, serial port, RS-232 port, IEEE-488 port, Bluetooth connectivity, wireless LAN connectivity, TCP / IP connectivity, Ethernet connectivity, control voltage interface, MIDI interface, analog input interface, and digital input interface.

[0119] As used herein, “display” or “display device” encompasses an output device or user interface suitable for displaying images or data. Displays can output visual, audio, and / or tactile data. Examples of displays include, but are not limited to: computer monitors, television screens, touchscreens, tactile electronic displays, Braille screens, cathode ray tubes (CRTs), memory tubes, bistable displays, electronic paper, vector displays, flat panel displays, vacuum fluorescent displays (VFs), light-emitting diode (LED) displays, electroluminescent displays (ELDs), plasma display panels (PDPs), liquid crystal displays (LCDs), organic light-emitting diode (OLED) displays, projectors, and head-mounted displays.

[0120] Although the invention has been described and illustrated in detail in the accompanying drawings and the foregoing description, such description and illustration are to be regarded as illustrative or exemplary rather than restrictive; the invention is not limited to the disclosed embodiments.

[0121] Those skilled in the art, through studying the accompanying drawings, description, and claims, will be able to understand and implement other variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality. A single processor or other unit may perform the functions of several items recited in the claims. Although specific elements are recited in dissimilar dependent claims, this does not indicate that combinations of these elements cannot be advantageously used. Computer programs may be stored and / or distributed on suitable media, such as optical storage media or solid-state media provided with or as part of other hardware, but computer programs may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. No reference numerals in the claims shall be construed as limiting the scope.

Claims

1. A method for detecting motion artifacts in an input image, comprising: - Using (201-203) to represent the k-space sampling mode and its timing k-space acquisition properties to generate a motion-damaged image with motion artifacts caused by a motion mode of a first initial setting type, such that the motion artifacts are defined as a function of the motion-damaged image and a feature matrix representing the phase-encoded damage caused by motion; - Initialize (205) at least one feature map using the value of the feature matrix, the at least one feature map representing the convolution kernel or multiplication kernel of the convolutional neural network; - Use training images to train (207) an initialized convolutional neural network (CNN), the training images being corrupted by a second training motion mode to obtain motion artifacts; - Use a trained CNN to obtain motion artifacts in the input image (209).

2. The method according to claim 1, wherein, The initialization includes determining, for each feature map of the CNN, a feature matrix with different first initial setting types for motion patterns, and using the values ​​of the feature matrix to initialize the feature map.

3. The method according to claim 1 or 2, wherein, The motion artifacts are defined in k-space as the product of the feature matrix and the motion-damaged image, wherein the feature map is a multiplication layer of the CNN, and the CNN is configured to receive a k-space representation of the image as input.

4. The method according to claim 1 or 2, wherein, The motion artifacts are defined in image space as the convolution of the feature matrix and the motion-damaged image, wherein the feature map is the kernel of the CNN, and the CNN is configured to receive the image as input.

5. The method according to claim 1 or 2, wherein the motion of the first initial setting type is a translational motion characterized by a corresponding translational displacement.

6. The method according to claim 1 or 2, wherein the second training movement pattern is a combination of the first initial movement patterns, wherein, Each initial motion pattern is characterized by a corresponding translational displacement.

7. The method according to claim 6, wherein, The motion artifacts in the input image are caused by motion, which is: -A combination of the first initial movement pattern and the second training movement pattern; or - The second training exercise mode.

8. The method according to claim 6, wherein, The first initial movement pattern may be different from or the same as the second training movement pattern.

9. The method according to claim 1 or 2, wherein the k-space acquisition properties include the sampling mode and / or sampling timing of the k-space.

10. The method according to claim 1 or 2, wherein, The motion artifact A of the motion-damaged image M is defined as the convolution of the motion-damaged image M and the feature matrix K, wherein the feature matrix K is composed of... Define, where, δ is the δ function, dx(t) is the motion function, FT represents the Fourier transform, and k is the k-space location of the motion-damaged image M.

11. The method according to claim 1 or 2, wherein, The motion artifact A of the motion-damaged image M in k-space is defined as the product of the motion-damaged image M in k-space and the feature matrix, wherein the feature matrix is ​​composed of... Define, where, δ is the δ function, dx(t) is the motion function, FT represents the Fourier transform, and k is the k-space location of the motion-damaged image M.

12. A method for training a convolutional neural network (CNN) for motion artifact detection, comprising: - Using (201-203) to represent the k-space sampling mode and its timing k-space acquisition properties to generate a motion-damaged image with motion artifacts caused by a motion mode of a first initial setting type, such that the motion artifacts are defined as a function of the motion-damaged image and a feature matrix representing the phase-encoded damage caused by motion; - Initialize (205) at least one feature map using the value of the feature matrix, the at least one feature map representing the convolution kernel or multiplication kernel of the convolutional neural network; - Use training images to train (207) an initialized convolutional neural network (CNN), the training images being corrupted by a second training motion mode to obtain motion artifacts.

13. A computer program product comprising machine-executable instructions for execution by a processor, wherein, The execution of the machine-executable instructions causes the processor to perform the method according to any one of the preceding claims.

14. A medical analysis system (111), the medical analysis system comprising: At least one processor (107); and at least one memory (103) including computer program code; The at least one memory (103) and the computer program code are configured to use the at least one processor (107) to cause the medical analysis system to perform at least the method according to any one of claims 1-12.

15. An MRI system (700) comprising a medical analysis system (111) according to claim 14, the MRI system (700) being configured to acquire image data and reconstruct an initial image from the image data, the initial image being processed by the medical analysis system according to claim 14 to perform the method according to any one of claims 1-12.

Images