System and method for enhancing propeller image quality by denoising a blade

Abstract

A system and method for improving image quality of periodic rotation overlapping parallel line with enhanced reconstruction (PROPELLER) imaging: during a PROPELLER sequence, acquiring a plurality of k-space data blades of a region of interest from a coil via a magnetic resonance imaging (MRI) scanner in a rotation about a center of k-space, wherein each blade of the plurality of k-space data blades comprises a plurality of parallel phase-coded lines sampled in a phase-coded order. The system and method also include denoising each of the plurality of k-space data leaves using a deep learning-based denoising network to generate a plurality of denoised leaves. The system and method also include generating a complex image from the plurality of denoised leaves using a PROPELLER reconstruction algorithm.

Description

Background Art

[0001] The subject matter disclosed herein relates to medical imaging, and more particularly to a system and method for enhancing PROPELLER image quality by denoising blades.

[0002] Non-invasive imaging techniques allow images of internal structures or features of a patient / object to be obtained without performing an invasive procedure on the patient / object. Specifically, such non-invasive imaging techniques rely on various physical principles (such as differential transmission of X-rays through a target volume, reflection of acoustic waves within a volume, paramagnetism of different tissues and materials within a volume, decomposition of target radionuclides within the body, etc.) to acquire data and construct an image or otherwise represent the observed internal features of the patient / object.

[0003] During magnetic resonance imaging (MRI), when a substance such as human tissue is subjected to a uniform magnetic field (polarization field B0), the individual magnetic moments of the spins in the tissue attempt to align with the polarization field, but precess around it in a random order at their characteristic Larmor frequency. If the substance or tissue is subjected to a magnetic field (excitation field B1) that is in the xy plane and close to the Larmor frequency, the net alignment moment or "longitudinal magnetization" M z can be rotated or "tilted" into the xy plane to produce a net transverse magnetic moment M t After the excitation signal B1 is terminated, a signal is emitted by the excited spins and can be received and processed to form an image.

[0004] When these signals are used to generate images, the magnetic field gradient (G x , G y and G z ). Typically, the area to be imaged is scanned in a series of measurement cycles in which these gradient fields are varied depending on the particular localization method being used. The resulting set of received nuclear magnetic resonance (NMR) signals is digitized and processed to reconstruct an image using one of a number of well-known reconstruction techniques.

[0005] Periodically Rotated Overlapping Parallel Lines with Enhanced Reconstruction (PROPELLER) is an MR technique that provides high-resolution magnetic resonance imaging with reduced motion artifacts by providing the ability to remove leaves affected by motion and by oversampling low spatial frequencies. Although the noise is Gaussian in each leaf, after the leaves are gridded into a Cartesian matrix, the noise becomes colored noise. Attempts to denoise colored noise have been reported to result in blurring and degradation of image quality. In the case of low signal-to-noise ratio data, the denoising performance becomes very poor. Summary of the invention

[0006] The following shows an overview of certain embodiments disclosed herein. It should be understood that these aspects are provided only to provide the reader with a brief overview of these specific embodiments, and these aspects are not intended to limit the scope of the present disclosure. In fact, the present disclosure may cover various aspects that may not be shown below.

[0007] In one embodiment, a computer-implemented method for improving the image quality of periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) imaging is provided. The computer-implemented method includes: acquiring, via a processor, a plurality of k-space data leaves of a region of interest from a coil in a manner rotating around the center of k-space via a magnetic resonance imaging (MRI) scanner during a PROPELLER sequence, wherein each of the plurality of k-space data leaves includes a plurality of parallel phase-encoded lines sampled in a phase-encoded order. The computer-implemented method also includes: denoising, via the processor, each of the plurality of k-space data leaves using a deep learning-based denoising network to generate a plurality of denoised leaves. The computer-implemented method also includes: generating, via the processor, a complex image from the plurality of denoised leaves using a PROPELLER reconstruction algorithm.

[0008] In one embodiment, the plurality of k-space data blades are acquired from a single channel of the coil.

[0009] In one embodiment, the plurality of k-space data blades are acquired from a plurality of channels of the coil.

[0010] In one embodiment, the computer-implemented method further comprises combining, via the processor, the plurality of k-space data leaves acquired from the plurality of channels before denoising each of the plurality of k-space data leaves using the deep learning-based denoising network.

[0011] In one embodiment, the computer-implemented method further comprises: removing streaks, via the processor, using a deep learning-based streak removal network on the complex image.

[0012] In one embodiment, the computer-implemented method further comprises: training, via the processor, the deep learning-based denoising network on input-output data pairs using supervised learning. The input-output data pairs include near-perfect and conventional MR images simulated from natural images. The deep learning-based denoising network is trained to predict noise in Cartesian acquired images.

[0013] In one embodiment, at least some of the simulated image pairs include skewed aspect ratios.

[0014] In another embodiment, a system for improving the image quality of periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) imaging is provided. The system includes a memory that encodes a processor executable routine. The system also includes a processor that is configured to access the memory and execute the processor executable routine, wherein the routine causes the processor to perform an action when executed by the processor. The action includes: during a PROPELLER sequence, a plurality of k-space data leaves of a region of interest are acquired from a coil in a manner rotating around the center of k-space via a magnetic resonance imaging (MRI) scanner, wherein each of the plurality of k-space data leaves includes a plurality of parallel phase encoding lines sampled in a phase encoding order. The action also includes: denoising each of the plurality of k-space data leaves using a deep learning-based denoising network to generate a plurality of denoised leaves. The action also includes: generating a complex image from the plurality of denoised leaves using a PROPELLER reconstruction algorithm.

[0015] In one embodiment, the plurality of k-space data blades are acquired from a single channel of the coil.

[0016] In one embodiment, the plurality of k-space data blades are acquired from a plurality of channels of the coil.

[0017] In one embodiment, the actions further include combining, via the processor, the plurality of k-space data leaves acquired from the plurality of channels before denoising each of the plurality of k-space data leaves using the deep learning based denoising network.

[0018] In one embodiment, the actions further include: removing streaks on the complex image using a deep learning based streak removal network, via the processor.

[0019] In one embodiment, the actions further include: training, via the processor, the deep learning-based denoising network on input-output data pairs using supervised learning. The input-output data pairs include near-perfect and conventional MR images simulated from natural images. The deep learning-based denoising network is trained to predict noise in Cartesian acquired images.

[0020] In one embodiment, at least some of the simulated image pairs include skewed aspect ratios.

[0021] In another embodiment, a non-transitory computer-readable medium includes processor executable code that, when executed by a processor, causes the processor to perform an action. The action includes: during a PROPELLER sequence, acquiring a plurality of k-space data leaves of a region of interest from a coil in a manner rotating around the center of k-space via a magnetic resonance imaging (MRI) scanner, wherein each of the plurality of k-space data leaves includes a plurality of parallel phase encoding lines sampled in a phase encoding order. The action also includes: denoising each of the plurality of k-space data leaves using a deep learning-based denoising network to generate a plurality of denoised leaves. The action also includes: generating a complex image from the plurality of denoised leaves using a PROPELLER reconstruction algorithm.

[0022] In one embodiment, the plurality of k-space data blades are acquired from a single channel of the coil.

[0023] In one embodiment, the plurality of k-space data blades are acquired from a plurality of channels of the coil.

[0024] In one embodiment, the actions further include combining, via the processor, the plurality of k-space data leaves acquired from the plurality of channels before denoising each of the plurality of k-space data leaves using the deep learning based denoising network.

[0025] In one embodiment, the actions further include: removing streaks on the complex image using a deep learning based streak removal network, via the processor.

[0026] In one embodiment, the actions further include: training, via the processor, the deep learning-based denoising network on input-output data pairs using supervised learning. The input-output data pairs include near-perfect and conventional MR images simulated from natural images. The deep learning-based denoising network is trained to predict noise in Cartesian acquired images.

[0027] In one embodiment, at least some of the simulated image pairs include skewed aspect ratios. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] These and other features, aspects, and advantages of the present subject matter will be better understood when the following detailed description is read with reference to the accompanying drawings, in which like characters refer to like parts throughout the several views, and in which:

[0029] Figure 1 An embodiment of a magnetic resonance imaging (MRI) system suitable for use with the disclosed technology is illustrated;

[0030] Figure 2A flow chart illustrating a method for improving image quality of Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) imaging (eg, using data from a single channel) according to aspects of the present disclosure;

[0031] Figure 3 A flow chart illustrating a method for improving image quality of Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) imaging (eg, utilizing data from multiple channels) according to aspects of the present disclosure;

[0032] Figure 4 A flow chart illustrating a method for improving image quality of Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) imaging (eg, using detailed information of PROPELLER reconstruction) in accordance with aspects of the present disclosure;

[0033] Figure 5 A flowchart illustrating a method for training a deep learning based denoising network according to aspects of the present disclosure is illustrated;

[0034] Figure 6 depicts Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) images of a phantom with and without blade-level denoising according to aspects of the present disclosure;

[0035] Figure 7 depicts a periodically rotated overlapping parallel lines enhanced reconstruction (PROPELLER) image of a knee without blade level denoising and with blade level denoising without re-adding residuals and with re-adding 15% residuals according to aspects of the present disclosure;

[0036] Figure 8 depicts Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) images of a knee with and without blade-level denoising and a denoised residual-only PROPELLER reconstructed image;

[0037] Fig. 9 depicts periodically rotated overlapping parallel lines partner enhanced reconstruction (PROPELLER) images of a knee with and without blade level denoising (e.g., by applying denoising to the blades and applying denoising after meshing the blades) in accordance with aspects of the present disclosure;

[0038] Fig.10 depicts a periodically rotated overlapping parallel lines partner enhanced reconstruction (PROPELLER) image of a knee with and without blade level denoising (e.g., by applying denoising to the blade and applying denoising after meshing the blade) and a denoised residual image in accordance with aspects of the present disclosure;

[0039] Fig.11 depicts periodically rotated overlapping parallel lines partner enhanced reconstruction (PROPELLER) images (e.g., zoomed images) of a knee with and without blade level denoising (e.g., by applying denoising to the blades and applying denoising after meshing the blades) in accordance with aspects of the present disclosure;

[0040] Fig.12 depicts a Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) image of a spine with and without blade-level denoising performed (with and without re-adding residuals prior to meshing) in accordance with aspects of the present disclosure;

[0041] Fig.13 depicts periodically rotated overlapping parallel lines partner enhanced reconstruction (PROPELLER) images of a spine with and without blade-level denoising (e.g., by applying denoising to the blades and applying denoising after meshing the blades) in accordance with aspects of the present disclosure; and

[0042] Fig.14 Depicted are Periodically Rotated Overlapping Parallel Lines Enhanced Reconstruction (PROPELLER) images of a knee with and without blade-level denoising (eg, using different types of denoising) in accordance with aspects of the present disclosure. DETAILED DESCRIPTION

[0043] One or more specific embodiments will be described below. In order to provide a concise description of these embodiments, not all features of an actual implementation are described in the specification. It should be understood that in the development of any such actual implementation, as in any engineering or design project, many implementation-specific decisions must be made to achieve the developer's specific goals, such as complying with system-related and business-related constraints that may vary from implementation to implementation. In addition, it should be understood that such development efforts may be complex and time-consuming, but are still routine tasks for design, fabrication, and manufacturing for ordinary technicians who benefit from this disclosure.

[0044] When introducing elements of various embodiments of the present subject matter, the articles "a," "an," "the," and "said" are intended to indicate that there are one or more elements. The terms "comprising," "including," and "having" are intended to be inclusive, and mean that there may be additional elements in addition to the listed elements. Furthermore, any numerical examples in the following discussion are intended to be non-limiting, and thus the additional numerical values, ranges, and percentages are within the scope of the disclosed embodiments.

[0045] Although various aspects of the following discussion are provided in the context of medical imaging, it should be understood that the disclosed technology is not limited to such medical contexts. In fact, the examples and explanations are provided in such medical contexts only to facilitate explanation by providing examples of real-life implementations and applications. However, the disclosed technology can also be used in other contexts, such as non-destructive inspection of manufactured parts or goods (i.e., quality control or quality review application scenarios) and / or image reconstruction of non-invasive inspection of packages, boxes, suitcases, etc. (i.e., security inspection or screening application scenarios). In general, the disclosed technology can be used in any imaging or screening context or image processing or photography field, where a group or class of acquired data undergoes a reconstruction process to generate an image or volume.

[0046] The deep learning (DL) methods discussed herein may be based on artificial neural networks and thus may encompass one or more of the following: deep neural networks, fully interconnected networks, convolutional neural networks (CNNs), unfolded neural networks, perceptrons, codecs, recurrent networks, transformer networks, wavelet filter banks, u-nets, generative adversarial networks (GANs), dense neural networks (e.g., residual dense networks (RDNs)), or other neural network architectures. Neural networks may include shortcuts, activations, batch normalization layers, and / or other features. These techniques are referred to herein as DL techniques, although the term may also be used with particular reference to the use of deep neural networks, which are neural networks with multiple layers.

[0047] As discussed herein, DL techniques (which may also be referred to as deep machine learning, hierarchical learning, or deep structured learning) are a branch of machine learning techniques that employ mathematical representations of data and artificial neural networks for learning and processing such representations. For example, DL methods may be characterized as using one or more algorithms to extract or model highly abstract concepts of a class of data of interest. This may be accomplished using one or more processing layers, where each layer typically corresponds to a different level of abstraction, and thus may employ or utilize different aspects of the initial data or the output of the previous layer (i.e., a hierarchical or cascaded structure of layers) as the target of the process or algorithm of a given layer. In the context of image processing or reconstruction, this may be characterized as different layers corresponding to different feature levels or resolutions in the data. In general, the processing of one representation space to the next level of representation space may be viewed as a "stage" of the process. Each stage of the process may be performed by a separate neural network or by different parts of a larger neural network.

[0048] In magnetic resonance imaging, the noise in the acquired complex k-space and in the complex image is Gaussian. Periodically Rotated Overlapping Parallel Lines with Enhanced Reconstruction (PROPELLER) is an MR technique that provides high-resolution magnetic resonance imaging with reduced motion artifacts by providing the ability to remove motion-affected blades and by oversampling low spatial frequencies. In PROPELLER, overlapping blades are acquired around the k-space rotation. Therefore, the noise remains Gaussian in the blades. However, in order to obtain the final image, the blades need to be phase corrected and then gridded into uniform k-space coordinates (i.e., a Cartesian matrix), which turns the noise into colored noise. Typically, denoising is performed after the complex image is generated (when the noise is colored). However, it has been observed that denoising models perform better for Gaussian noise than for colored noise. It has been reported that attempts to denoise colored noise result in blurring and degradation of image quality. In the case of low signal-to-noise ratio data, denoising performance becomes very poor.

[0049] The present disclosure provides a system and method for improving the image quality of PROPELLER imaging. Specifically, the present disclosure provides a method for denoising each k-space data leaf before phase correction and gridding (wherein the noise is still Gaussian noise).

[0050] The disclosed systems and methods include: during a PROPELLER sequence, acquiring a plurality of k-space data leaves of a region of interest from a coil in a rotational manner around a center of k-space via a magnetic resonance imaging (MRI) scanner, wherein each of the plurality of k-space data leaves includes a plurality of parallel phase encoding lines sampled in a phase encoding order. The disclosed systems and methods also include: denoising each of the plurality of k-space data leaves using a deep learning-based denoising network to generate a plurality of denoised leaves. The disclosed systems and methods also include: generating a complex image from the plurality of denoised leaves using a PROPELLER reconstruction algorithm. In certain embodiments, the plurality of k-space data leaves are acquired from a single channel of the coil. In certain embodiments, the plurality of k-space data leaves are acquired from a plurality of channels of the coil. In these embodiments, the disclosed systems and methods include: combining the plurality of k-space data leaves acquired from the plurality of channels before denoising each of the plurality of k-space data leaves using a deep learning-based denoising network. In certain embodiments, the disclosed systems and methods also include: removing streaks on the complex image using a deep learning-based streak removal network. In certain embodiments, the disclosed systems and methods further include: applying super-resolution to the complex image to generate a higher resolution complex image that is further denoised (e.g., without blur). In certain embodiments, the complex image may be subjected to other post-processing. In certain embodiments, the disclosed systems and methods further include: writing the image to a DICOM file (e.g., for viewing or storage).

[0051] In certain embodiments, the disclosed systems and methods include: training a deep learning based denoising network on input-output data pairs using supervised learning. The input-output data pairs include near-perfect and conventional MRI-like images simulated from natural images. The deep learning based denoising network is trained to predict noise in Cartesian acquired images. In certain embodiments, at least some of the simulated image pairs have skewed aspect ratios to improve the denoising function of the deep learning based denoising network.

[0052] PROPELLER is typically used to suppress motion. Additionally, to suppress motion, PROPELLER can be accelerated, but this reduces the signal-to-noise ratio. Furthermore, if the imaging system is further degraded (with respect to the number of coils and field strength utilized), the signal-to-noise ratio drops below diagnostic quality. The disclosed system and method improves image quality in both signal-to-noise ratio and clarity.

[0053] Considering the above, Figure 1, a magnetic resonance imaging (MRI) system 100 is schematically illustrated as including a scanner 102, scanner control circuitry 104, and system control circuitry 106. According to embodiments described herein, the magnetic resonance imaging system 100 is generally configured to perform magnetic resonance imaging.

[0054] The system 100 also includes a remote access and storage system or device, such as a picture archiving and communication system (PACS) 108, or other devices, such as teleradiology equipment, that enable on-site or off-site access to data acquired by the system 100. In this way, MR data can be acquired and then processed and evaluated on-site or off-site. Although the magnetic resonance imaging system 100 can include any suitable scanner or detector, in the illustrated embodiment, the system 100 includes a whole body scanner 102 having a housing 120 through which an aperture 122 is formed. An examination table 124 can be moved into the aperture 122 to allow a patient 126 (e.g., a subject) to be positioned therein to image selected anatomical structures within the patient's body.

[0055] The scanner 102 includes a series of associated coils for generating a controlled magnetic field for exciting gyromagnetic materials within the anatomical structure of the patient being imaged. Specifically, a primary magnetic coil 128 is provided for generating a primary magnetic field B0 that is generally aligned with the orifice 122. A series of gradient coils 130, 132, and 134 allow controlled gradient magnetic fields to be generated during an examination sequence for position encoding of certain gyromagnetic nuclei within the patient 126. A radio frequency (RF) coil 136 (e.g., an RF transmit coil) is configured to generate radio frequency pulses for exciting certain gyromagnetic nuclei within the patient. In addition to the coils that may be located locally at the scanner 102, the system 100 also includes a set of receiving coils or RF receiving coils 138 (e.g., a coil array) configured to be placed proximal to the patient 126 (e.g., against the patient). For example, the receiving coil 138 may include a cervical / thoracic / lumbar (CTL) coil, a head coil, a single-sided spine coil, etc. Generally, the receive coil 138 is placed near or on top of the patient 126 to receive the weak RF signals (weak relative to the transmit pulses generated by the scanner coil) generated by certain gyromagnetic nuclei in the patient's body when the patient 126 returns to his or her relaxed state.

[0056] The various coils of the system 100 are controlled by external circuitry to generate the required fields and pulses and to read the emissions from the gyromagnetic material in a controlled manner. In the illustrated embodiment, the main power supply 140 provides power to the primary field coil 128 to generate the main magnetic field B0. The power input (e.g., power from a utility or grid), a power distribution unit (PDU), a power supply (PS), and a drive circuit 150 can provide power together to pulse the gradient field coils 130, 132, and 134. The drive circuit 150 may include an amplification and control circuit system for supplying current to the coils as defined by the digitized pulse sequence output by the scanner control circuit system 104.

[0057] Another control circuit 152 is provided for regulating the operation of the RF coil 136. The circuit 152 includes a switching device for alternating between an active mode of operation and a passive mode of operation, wherein the RF coil 136 transmits and does not transmit a signal, respectively. The circuit 152 also includes an amplification circuit system configured to generate RF pulses. Similarly, the receiving coil 138 is connected to a switch 154 that is capable of switching the receiving coil 138 between a receiving mode and a non-receiving mode. Thus, in the receiving mode, the receiving coil 138 resonates with the RF signal generated by the release of the gyromagnetic nuclei in the patient 126, and in the non-receiving mode, they do not resonate with the RF energy from the transmitting coil (i.e., the coil 136) to prevent undesired operation. In addition, the receiving circuit 156 is configured to receive data detected by the receiving coil 138, and may include one or more multiplexing and / or amplification circuits.

[0058] It should be noted that although the scanner 102 and control / amplification circuit system described above are illustrated as being coupled by a single line, in actual examples there may be many such lines. For example, separate lines may be used for control, data communication, power transmission, etc. In addition, appropriate hardware may be provided along each type of line for properly processing data and current / voltage. In practice, various filters, digitizers, and processors may be provided between the scanner and either or both of the scanner control circuit 104 and the system control circuit system 106.

[0059] As shown, the scanner control circuit system 104 includes an interface circuit 158 ​​that outputs signals for driving the gradient field coils and the RF coils and for receiving data representing magnetic resonance signals generated in the examination sequence. The interface circuit 158 ​​is coupled to a control and analysis circuit 160. Based on a defined scheme selected via the system control circuit 106, the control and analysis circuit 160 executes commands for driving the circuits 150 and 152.

[0060] The control and analysis circuitry 160 is also used to receive magnetic resonance signals and perform subsequent processing before transmitting the data to the system control circuitry 106. The scanner control circuitry 104 also includes one or more memory circuits 162 that store configuration parameters, pulse sequence descriptions, examination results, etc. during operation.

[0061] Interface circuit 164 is coupled to control and analysis circuit 160 for exchanging data between scanner control circuit system 104 and system control circuit system 106. In some embodiments, control and analysis circuit 160, although shown as a single unit, may include one or more hardware devices. System control circuit 106 includes interface circuit 166, which receives data from scanner control circuit system 104 and transmits data and commands back to scanner control circuit system 104. Control and analysis circuit 168 may include a CPU in a general or special purpose computer or workstation. Control and analysis circuit 168 is coupled to memory circuit 170 to store programming code for operating magnetic resonance imaging system 100, and to store processed image data for later reconstruction, display and transmission. The programming code may execute one or more algorithms that are configured to perform reconstruction of acquired data as described below when executed by a processor. In some embodiments, memory circuit 170 may store one or more neural networks (e.g., a denoising network based on deep learning, a de-striping network based on deep learning, etc.) for processing and / or reconstructing acquired data, as described below. In certain embodiments, image reconstruction may occur on a separate computing device having processing circuitry and memory circuitry.

[0062] The processing components (e.g., microprocessors or processing circuitry) and memory (such as may be present in the scanner control circuitry 104 and / or the system control circuitry 106) of the magnetic resonance imaging system 100 may be used to execute stored software code, instructions or routines for acquiring and processing MR data. As used herein, the term "code" or "software code" refers to any instructions or instruction sets for controlling the magnetic resonance imaging system 100. The code or software code may be in a computer executable form, such as machine code, which is a set of instructions and data that is directly executed by the processing components of the scanner control circuitry 104 and / or the system control circuitry 106; a human understandable form, such as source code, which may be compiled for execution by the processing components of the scanner control circuitry 104 and / or the system control circuitry 106; or an intermediate form, such as object code, which is generated by a compiler. In some embodiments, the magnetic resonance imaging system 100 may include multiple controllers.

[0063] For example, the memory may store processor executable software code or instructions (e.g., firmware or software) that are tangibly stored on a non-transitory computer readable medium. Additionally or alternatively, the memory 46 may store data. For example, the memory may include volatile memory (such as random access memory (RAM)) and / or non-volatile memory (such as read-only memory (ROM), flash memory, hard drive, or any other suitable optical, magnetic or solid-state storage medium or a combination thereof. In addition, the processing unit may include multiple microprocessors, one or more "general" microprocessors, one or more special-purpose microprocessors, and / or one or more application-specific integrated circuits (ASICS) or some combination thereof. For example, the processing unit may include one or more reduced instruction set (RISC) or complex instruction set (CISC) processors. The processing unit may include multiple processors and / or the memory may include multiple memory devices.

[0064] In some embodiments, the processor may be configured to: during a PROPELLER sequence, acquire multiple k-space data leaves of a region of interest from a coil in a manner rotating around the center of k-space via a magnetic resonance imaging (MRI) scanner, wherein each of the multiple k-space data leaves includes multiple parallel phase encoding lines sampled in a phase encoding order. The processor may also be configured to: denoise each of the multiple k-space data leaves using a deep learning-based denoising network to generate multiple denoised leaves. The processor may also be configured to: generate a complex image from the multiple denoised leaves using a PROPELLER reconstruction algorithm. In some embodiments, in the case where multiple k-space data leaves are acquired from multiple channels of the coil, the processor may even be configured to: combine the multiple k-space data leaves acquired from the multiple channels before denoising each of the multiple k-space data leaves using a deep learning-based denoising network. In some embodiments, the processor may also be configured to: remove stripes on the complex image using a deep learning-based stripe removal network. In some embodiments, the processor may apply super-resolution to the complex image to generate a higher resolution complex image that is further denoised (e.g., without blurring). In some embodiments, the complex image may be subjected to other post-processing. In some embodiments, the processor may write the image to a DICOM file (e.g., for viewing or storage).

[0065] In some embodiments, the processor may be further configured to: train a deep learning based denoising network on input-output data pairs using supervised learning. The input-output data pairs include near-perfect and conventional MR images simulated from natural images. The deep learning based denoising network is trained to predict noise in Cartesian acquired images. In some embodiments, at least some of the simulated image pairs have skewed aspect ratios to improve the denoising function of the deep learning based denoising network.

[0066] Additional interface circuitry 172 may be provided for exchanging image data, configuration parameters, etc. with external system components (such as remote access and storage device 108). Finally, system control and analysis circuitry 168 may be communicatively coupled to various peripheral devices for facilitating operator interface and generating hard copies of reconstructed images. In the illustrated embodiment, these peripheral devices include a printer 174, a monitor 176, and a user interface 178, which includes devices such as a keyboard, a mouse, a touch screen (e.g., integral to the monitor 176), and the like.

[0067] Figure 2 is a flow chart of a method 180 for improving image quality of PROPELLER imaging. One or more steps of the method 180 may be performed by Figure 1 The processing circuitry of the magnetic resonance imaging system 100 or a remote computing system may be performed. For example, the processing circuitry may be part of the scanner control circuitry 104 and / or the system control circuitry 106 of the magnetic resonance imaging system 100. One or more steps of the method 180 may be performed simultaneously and / or in Figure 2 The different orders shown perform one or more steps of the method.The method 180 is performed on data acquired from a single channel of a coil (eg, an RF receive coil such as a body coil).

[0068] Method 180 includes: during a PROPELLER sequence, via a magnetic resonance imaging scanner (e.g., Figure 1A magnetic resonance imaging scanner 102 in the apparatus 100 acquires a plurality of k-space data leaves of a region of interest from a coil (e.g., an RF receive coil such as a body coil) in a rotational manner about a center of k-space (e.g., rotating approximately 10 to 20 degrees between leaf acquisitions) (box 182). Each of the plurality of k-space data leaves includes a plurality of parallel phase encoding lines (e.g., having a straight line shape) sampled in a phase encoding order using a fast spin echo method or a gradient echo method (i.e., each leaf is filled by an echo chain of a corresponding MR pulse sequence). The method 180 also includes: denoising each of the plurality of k-space data leaves (e.g., individually) using a deep learning-based denoising network or model to generate a plurality of denoised leaves (box 184). The deep learning-based denoising network is trained to predict noise present in a Cartesian acquired image (e.g., an MR image). The deep learning-based denoising network is constructed to be size-invariant. As described below Figure 5 The method 180 also includes: using the PROPELLER reconstruction algorithm (as described below in Figure 4 A complex image of the region of interest is generated from the multiple de-noised leaves (block 186) as described in . For example, the PROPELLER reconstruction algorithm includes: phase correction for each leaf to ensure that its rotation point is precisely located at the center of k-space; correction for the object's overall in-plane rotation and in-plane translation; and correlation weighting to minimize data from leaves that contain motion or displacement errors. In certain embodiments, the complex image is further processed (as described below in ) before writing the image to a Digital Imaging and Communications in Medicine (DICOM) file. Figure 4 ).

[0069] Figure 3 1 is a flow chart of a method 188 for improving image quality of PROPELLER imaging (eg, using data from multiple channels). One or more steps of the method 188 may be performed by Figure 1 The processing circuitry of the magnetic resonance imaging system 100 or a remote computing system may be performed. For example, the processing circuitry may be part of the scanner control circuitry 104 and / or the system control circuitry 106 of the magnetic resonance imaging system 100. One or more steps of the method 188 may be performed simultaneously and / or in Figure 3 One or more steps of the method are performed in different orders as shown.The method 188 is performed on data acquired from a plurality of channels of a coil (eg, an RF receive coil such as a body coil).

[0070] Method 188 includes: during a PROPELLER sequence, via a magnetic resonance imaging scanner (e.g., Figure 1The method 188 further includes: combining the plurality of k-space data leaves acquired from the plurality of channels to denoise each of the plurality of k-space data leaves (i.e., combining corresponding k-space data leaves from the channels) in a phase encoding order using a fast spin echo method or a gradient echo method (i.e., each leaf is filled by an echo chain of a corresponding MR pulse sequence) before denoising each of the plurality of k-space data leaves using a deep learning-based denoising network.

[0071] The method 188 also includes: denoising each leaf (e.g., a combined leaf) in the plurality of k-space data leaves using a deep learning-based denoising network or model to generate a plurality of denoised leaves (block 194). The deep learning-based denoising network is trained to predict the noise present in a Cartesian acquisition image (e.g., an MR image). The deep learning-based denoising network is constructed to be size-invariant. Figure 5 The method 188 also includes: using the PROPELLER reconstruction algorithm (as described below in Figure 4 A complex image of the region of interest is generated from the plurality of de-noised leaves (block 196) as described in . For example, the PROPELLER reconstruction algorithm includes: phase correction for each leaf to ensure that its rotation point is precisely centered in k-space; correction for global in-plane rotation and in-plane translation of the object; and correlation weighting to minimize data from leaves that contain motion or displacement errors. In certain embodiments, the complex image is further processed (as described below in ) before writing the image to a Digital Imaging and Communications in Medicine (DICOM) file. Figure 4 ).

[0072] Figure 4 is a flow chart of a method 198 for improving image quality of PROPELLER imaging (eg, by detailed information of PROPELLER reconstruction). One or more steps of the method 198 may be performed by Figure 1 The processing circuitry of the magnetic resonance imaging system 100 or a remote computing system may be performed. For example, the processing circuitry may be part of the scanner control circuitry 104 and / or the system control circuitry 106 of the magnetic resonance imaging system 100. One or more steps of the method 198 may be performed simultaneously and / or in Figure 4One or more steps of the method may be performed in a different order than shown. Figure 2 ) or multiple channels (e.g., as described in method 180 in Figure 3 Execute method 198 based on the data collected (as described in method 188).

[0073] Method 198 includes: during a PROPELLER sequence, via a magnetic resonance imaging scanner (e.g., Figure 1 A magnetic resonance imaging scanner 102 in the apparatus acquires a plurality of k-space data leaves (e.g., from a single channel or a plurality of channels) of a region of interest from a coil (e.g., an RF receive coil such as a body coil) in a rotational manner about the center of k-space (e.g., rotating about 10 to 20 degrees between leaf acquisitions) (block 200). Each leaf of the plurality of k-space data leaves includes a plurality of parallel phase encoding lines (e.g., having a straight line shape) sampled in a phase encoding order using a fast spin echo method or a gradient echo method (i.e., each leaf is filled by an echo train of a corresponding MR pulse sequence).

[0074] The method 198 also includes: denoising each leaf (or each combined leaf) in the plurality of k-space data leaves using a deep learning-based denoising network or model to generate a plurality of denoised leaves (block 202). The deep learning-based denoising network is trained to predict the noise present in a Cartesian acquired image (e.g., an MR image). The deep learning-based denoising network is constructed to be size-invariant. Figure 5 The training of a deep learning based denoising network is described in.

[0075] As noted above, the PROPELLER reconstruction algorithm includes: phase correction for each leaf to ensure that its rotation point is precisely located at the center of k-space; correction for the object's overall in-plane rotation and in-plane translation; and correlation weighting to minimize data from leaves containing motion or displacement errors. Method 198 also includes: performing phase correction on multiple denoised leaves (box 204). Performing phase correction on each leaf ensures that its rotation point is precisely located at the center of k-space. Method 198 still also includes: determining a weight for each leaf and estimating a sampling density correction for each leaf (box 206). Based on the weights and the estimated sampling density correction to be applied to the denoised leaves, method 198 includes: gridding all denoised leaves in k-space into a Cartesian matrix (e.g., uniform k-space coordinates) (box 208). Boxes 208 and 210 of method 198 ensure correction for the object's overall in-plane rotation and in-plane translation, while correlation weighting minimizes data from leaves containing motion or displacement errors. The method 198 further includes generating a complex image from the gridded blade (block 210).

[0076] After the PROPELLER reconstruction, in some embodiments, the method 198 includes: removing streaks using a deep learning based de-streaking network or model on the complex image (box 212). The de-streaking network is trained to remove streaks that may cause artifacts in the image. In some embodiments, the method 198 also includes: applying super-resolution to the complex image to generate a higher resolution complex image that is further denoised (e.g., without blur) (box 214). In some embodiments, the complex image may be subjected to other post-processing. The method 198 also includes: writing the image to a DICOM file (e.g., for viewing or storage) (box 216).

[0077] Figure 5 A flow chart of a method 218 for training a deep learning based denoising network is illustrated. One or more steps of the method 218 may be performed by Figure 1 The processing circuitry of the magnetic resonance imaging system 100 or a remote computing system may be performed. For example, the processing circuitry may be part of the scanner control circuitry 104 and / or the system control circuitry 106 of the magnetic resonance imaging system 100.

[0078] The method 218 includes inputting input-output (e.g., labeled) data pairs 220 into a neural network 222 (box 224). In some embodiments, the neural network 222 is a residual dense network. In some embodiments, the input-output data pairs 220 include near-perfect (i.e., clear, artifact-free, and high signal-to-noise ratio) and conventional MRI-like images 225 simulated from natural images 223. The method 218 also includes training the neural network 222 on the input-output data pairs 220 using supervised learning to generate a deep learning-based denoising network or model 227 (box 228). In some embodiments, at least some of the input-output data pairs 220 include skewed aspect ratios to enhance training to improve the performance of the deep learning-based denoising network 227. The deep learning-based denoising network 227 is trained to predict noise in Cartesian acquired images (e.g., MR images). Since natural images with sufficient variability are trained and the residuals to be predicted remain in the same Gaussian distribution, deep learning based can be utilized on leaf level images. In some embodiments, the image of the input-output data pair 220 can be 256×256. However, the deep learning based denoising network 227 is constructed to be size invariant.

[0079] Figures 6 to 14 The effectiveness of performing leaf-level denoising is demonstrated. Figure 6PROPELLER images of a phantom with and without blade denoising are depicted. Images 230, 232, 234, and 236 were acquired from a phantom using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil. Images 230 and 234 are PROPELLER images without denoising. Images 232 and 236 are corresponding PROPELLER images with blade-level denoising (e.g., 100% denoising) performed using the deep learning-based denoising network described above. The quality (e.g., signal-to-noise ratio and clarity) of images 232 and 236 is improved compared to images 230 and 234.

[0080] Figure 7 PROPELLER images of a knee without blade-level denoising and with blade-level denoising without re-adding residuals and re-adding 15% residuals are depicted. Images 238, 240, and 242 were acquired from a knee using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via proton density weighted fat suppression scanning). Image 238 is a PROPELLER image without denoising. Image 240 is a corresponding PROPELLER image with blade-level denoising (e.g., 100% denoising) performed using the above-mentioned deep learning-based denoising network. Image 242 is a corresponding PROPELLER image with blade-level denoising performed using the above-mentioned deep learning-based denoising network but with 15% residuals re-added to re-add texture before meshing. The quality (e.g., signal-to-noise ratio and clarity) of images 240 and 242 is improved compared to image 238.

[0081] Figure 8 PROPELLER images of the knee with and without blade-level denoising and residual-only PROPELLER reconstructed images after denoising are depicted. Images 244, 246, and 248 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via proton density-weighted fat suppression scanning). Image 244 is a PROPELLER image without denoising. Image 246 is a corresponding PROPELLER image with blade-level denoising (e.g., 100% denoising) performed using the above-mentioned deep learning-based denoising network. Image 248 is a residual-only reconstructed PROPELLER image of all blades after denoising (intensity magnification). The quality (e.g., signal-to-noise ratio and clarity) of image 246 is improved compared to image 244. Image 248 verifies that individual denoising of the blades does not introduce artifacts.

[0082] Fig. 9PROPELLER images of a knee with and without blade-level denoising (e.g., by applying denoising to the blades and applying denoising after the blades are gridded) are depicted. Images 250, 252, and 254 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via a proton density weighted fat suppression scan). Image 250 is a PROPELLER image without denoising. Image 252 is a corresponding PROPELLER image in which blade-level denoising (e.g., 100% denoising) was performed using the deep learning-based denoising network described above. Image 254 is a corresponding PROPELLER image in which a gridded image was denoised using a specially trained residual dense network (e.g., on natural images after training) to denoise colored noise found after gridding Gaussian noise. The quality of image 252 is improved relative to images 250 and 254. In addition, image 252 is clearer than image 254. In addition, minute structures that are blurred out in image 254 are preserved in image 252. In other words, excessive blurring is observed in image 254 compared to image 252.

[0083] Fig.10 PROPELLER images of a knee with and without blade-level denoising (e.g., by applying denoising to the blades and applying denoising after the blades are gridded) and denoised residual images are depicted. Images 256, 258, and 260 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via proton density weighted fat suppression scanning). Image 256 is a PROPELLER image without denoising. Image 258 is a corresponding PROPELLER image with blade-level denoising (e.g., 100% denoising) performed using the above-mentioned deep learning-based denoising network. Image 256 is a corresponding PROPELLER image in which a gridded image is denoised using a specially trained residual dense network (e.g., on natural images after training) to denoise the colored noise found after gridding Gaussian noise. Image 262 is a denoised residual image between image 256 and image 258. Image 264 is the denoised residual image between image 256 and image 260. The structure present in image 264 indicates that the edges are blurred in the grid-level deep learning noise.

[0084] Fig.11PROPELLER images (e.g., enlarged images) of a knee with and without blade-level denoising (e.g., by applying denoising to the blades and applying denoising after the blades are gridded) are depicted. Images 266, 268, and 270 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via proton density weighted fat suppression scanning). Image 266 is an enlarged PROPELLER image that has not been denoised. Image 268 is an enlarged corresponding PROPELLER image that has been blade-level denoising (e.g., 100% denoising) performed using the above-mentioned deep learning-based denoising network. Image 270 is a (enlarged) corresponding PROPELLER image that denoises the gridded image using a specially trained residual dense network (e.g., trained natural images) to denoise the colored noise found after gridding Gaussian noise. Stripes are present in both image 266 (which is the source image) and image 268. However, the stripes stand out more in image 268 than in image 266 because the high noise in the source image masks the stripes. As depicted in image 270, the residual dense network used for grid-level denoising aggressively denoises at this high noise level, causing the stripes to be blurred out along with the edges and noise.

[0085] Fig.12 PROPELLER images of the spine with and without leaf-level denoising performed (with and without re-adding residuals before meshing). Images 272, 274, and 276 were acquired from the spine using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via a C-spine T2 scan). Image 272 is a PROPELLER image without denoising. Image 274 is a corresponding PROPELLER image with leaf-level denoising (e.g., 100% denoising) performed using the deep learning-based denoising network described above. Image 276 is a corresponding PROPELLER image with leaf-level denoising performed using the deep learning-based denoising network described above but with 15% of residuals re-added before meshing to re-add texture. The quality (e.g., signal-to-noise ratio and clarity) of images 274 and 276 is improved compared to image 272.

[0086] Fig.13PROPELLER images of the spine with and without leaf-level denoising (e.g., by applying denoising to the blades and applying denoising after the blades are gridded) are depicted. Images 278, 280, and 282 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (via a C-spine T2 scan). Image 278 is a PROPELLER image without denoising. Image 280 is a corresponding PROPELLER image in which leaf-level denoising (e.g., 100% denoising) was performed using the deep learning-based denoising network described above. Image 282 is a corresponding PROPELLER image in which a gridded image was denoised using a specially trained residual dense network (e.g., on natural images after training) to denoise colored noise found after gridding Gaussian noise. The quality of image 280 is improved relative to images 278 and 282. In addition, image 280 is clearer than image 282. In addition, minute structures that are blurred out in image 282 are preserved in image 280. In other words, excessive blurring is observed in image 282 compared to image 280.

[0087] Fig.14 PROPELLER images of the knee with and without blade-level denoising (e.g., using different types of denoising) are depicted. Images 284, 286, and 288 were acquired using PROPELLER using a 1.5 Tesla (T) magnetic resonance imaging scanner using a body coil (using a proton density weighted fat suppression scan). Image 284 is a PROPELLER image without denoising. Image 286 is a corresponding PROPELLER image with blade-level denoising (e.g., 85% denoising) performed using the deep learning-based denoising network described above. Image 288 is a PROPELLER image acquired using the Air Force 3D image processing software that performs denoising, de-streaking, and super-resolution. TM Corresponding PROPELLER images obtained with ReconDL (ARDL) PROPELLER. Compared to image 286, the denoising in image 288 is patchy and incomplete.

[0088] The technical effects of the disclosed subject matter include providing systems and methods for improving image quality in PROPELLER imaging. Specifically, the present disclosure provides denoising of individual k-space data leaves before phase correction and gridding of the k-space data leaves (where the noise is still Gaussian noise). The technical effects of the disclosed subject matter include improving image quality in terms of both signal-to-noise ratio and clarity.

[0089] The technology presented and claimed herein is referred to and applied to physical and specific examples of a practical nature that clearly improves upon the present state of the art and, therefore, is not abstract, intangible, or purely theoretical. In addition, if any claim appended to the end of this specification contains one or more elements designated as "means for [performing] the function of ..." or "steps for [performing] the function of ...," it is intended that such elements be interpreted under 35 U.S.C. § 112(f). However, for any claim containing elements designated in any other manner, it is not intended that such elements be interpreted under 35 U.S.C. § 112(f).

[0090] This written description uses examples to disclose the subject matter, including the best mode, and also to enable those skilled in the art to practice the subject matter, including making and using any devices or systems and performing any included methods. The patentable scope of the subject matter is defined by the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insignificant differences from the literal language of the claims.

Claims

1. A method for improving the image quality of PROPELLER imaging, the method comprising: Acquiring, via the processor, a plurality of k-space data leaves of a region of interest from a coil in a manner rotating about a center of k-space via the MRI scanner during a PROPELLER sequence, wherein each leaf of the plurality of k-space data leaves comprises a plurality of parallel phase encoding lines sampled in a phase encoding order; De-noising each of the plurality of k-space data leaves using a deep learning-based de-noising network via the processor to generate a plurality of de-noised leaves; as well as A complex image is generated from the plurality of de-noised blades using a PROPELLER reconstruction algorithm via the processor. The method of claim 1 , wherein the plurality of k-space data blades are acquired from a single channel of the coil. The method of claim 1 , wherein the plurality of k-space data blades are acquired from a plurality of channels of the coil.

4. The method according to claim 3, further comprising: Prior to denoising each of the plurality of k-space data leaves using the deep learning based denoising network, the plurality of k-space data leaves acquired from the plurality of channels are combined via the processor.

5. The method according to claim 1, further comprising: The processor removes streaks from the complex image using a deep learning based streak removal network.

6. The method according to claim 1, further comprising: The deep learning-based denoising network is trained, via the processor, on input-output data pairs using supervised learning, wherein the input-output data pairs include near-perfect and regular MR images simulated from natural images, and wherein the deep learning-based denoising network is trained to predict noise in Cartesian acquired images.

7. The method of claim 6, wherein at least some of the simulated image pairs include skewed aspect ratios.

8. A system for improving the image quality of PROPELLER imaging, the system comprising: a memory encoding a processor-executable routine; and A processor configured to access the memory and execute the processor-executable routine, wherein the routine, when executed by the processor, causes the processor to implement the method according to any one of claims 1 to 7.

9. A non-transitory computer readable medium comprising processor executable code, which when executed by a processor causes the processor to perform the method according to any one of claims 1 to 7.

Images