System and method for improving image quality of magnetically sensitive weighted imaging using deep learning
Deep learning technology to reduce noise and super-resolution processing of SWI images, solve the noise and signal-to-noise ratio problems in SWI technology, improve image quality and diagnostic value, and present clear blood vessel and tissue structure.
Patent Information
- Application Number
- CN202410482759.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-08
- Filing Date
- 2024-04-22
- Publication Date
- 2024-11-08
AI Technical Summary
Magnetic sensitive weighted imaging (SWI) technology is susceptible to noise during image reconstruction, resulting in a decrease in signal-to-noise ratio and blurred image, making it difficult to effectively distinguish between paramagnetic and resistant magnetic substances, especially in multi-echo situations, which may be reduced after noise reduction. Artifacts and excessive blur.
Denoising complex images is used to denoiser based on deep learning, and combined with super-resolution technology, the phase and amplitude images of multiple complex echoes acquired during the SWI sequence are processed to generate clear images with higher resolution. Reduce noise impact and improve signal-to-noise ratio.
Through deep learning noise reduction and super-resolution processing, the quality of SWI images is significantly improved, artifacts are reduced, signal-to-noise ratio is enhanced, and the blood vessels and tissue structures are presented, improving the naturalness and diagnostic value of the image.
Smart Images

Figure CN118914948A_ABST
Abstract
Description
Background Art
[0001] The subject matter disclosed herein relates to medical imaging, and more particularly, to a system and method for improving the image quality of susceptibility weighted imaging (SWI) using deep learning.
[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 according to the particular localization method 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] SWI is a high-resolution MRI sequence that is particularly sensitive to compounds that distort the local magnetic field, making it useful for detecting blood products, iron, and diamagnetic calcium. In addition, SWI is used as part of brain MR imaging. The phase image is high-pass filtered to eliminate slowly changing magnetic susceptibility changes. This is important for distinguishing paramagnetic and diamagnetic substances. When this filtered phase image is used to accentuate the signal loss directly observed in the amplitude image, it is usually boosted to a higher power and the noise is amplified. This produces an undesirable effect on the SWI image. 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 susceptibility weighted imaging (SWI) is provided. The computer-implemented method includes acquiring, by a processor, a complex signal of magnetic resonance imaging (MRI) having multiple complex echoes of a region of interest from a coil by an MRI scanner during a SWI sequence. The computer-implemented method also includes phase filtering, by a processor, each of the multiple complex echoes. The computer-implemented method also includes generating, by a processor, a corresponding phase image and a corresponding amplitude image from each phase-filtered complex echo. The computer-implemented method even includes combining, by a processor, the corresponding amplitude images of the multiple complex echoes with each other to generate a combined amplitude image, and combining the corresponding phase images of the multiple complex echoes with each other to generate a combined phase image. The computer-implemented method also includes generating, by a processor, a complex image from both the combined amplitude image and the combined phase image. The computer-implemented method also includes denoising, by a processor, the complex image using a denoising network based on deep learning to generate a denoised complex image.
[0008] In another embodiment, a system for improving the image quality of susceptibility weighted imaging (SWI) 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 acquiring a complex magnetic resonance imaging (MRI) signal having multiple echoes of a region of interest from a coil by an MRI scanner during a SWI sequence. The action also includes phase filtering each complex echo in the multiple echoes. The action also includes generating a corresponding phase image and a corresponding amplitude image from each complex echo that has been phase filtered. The action even further includes combining the corresponding amplitude images of the multiple complex echoes with each other to generate a combined amplitude image, and combining the corresponding phase images of the multiple complex echoes with each other to generate a combined phase image. The action also includes generating a complex image from both the combined amplitude image and the combined phase image. The action also includes denoising the complex image using a denoising network based on deep learning to generate a denoised complex image.
[0009] In another embodiment, a non-transitory computer-readable medium includes processor executable code, which causes the processor to perform an action when executed by the processor. The action includes acquiring a magnetic resonance imaging (MRI) complex signal with multiple echoes of a region of interest from a coil (e.g., a radio frequency receiving coil) by an MRI scanner during a multi-echo acquisition imaging sequence. The action also includes phase filtering each complex echo in the multiple echoes. The action also includes generating a corresponding phase image and a corresponding amplitude image from each phase-filtered complex echo. The action even further includes combining the corresponding amplitude images of the multiple complex echoes with each other to generate a combined amplitude image, and combining the corresponding phase images of the multiple complex echoes with each other to generate a combined phase image. The action also includes generating a complex image from both the combined amplitude image and the combined phase image. The action also includes denoising the complex image using a denoising network based on deep learning to generate a denoised complex image. The action also includes applying super-resolution to the denoised complex image to generate a fully denoised complex image of a higher resolution without any blur. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] 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:
[0011] Figure 1 An embodiment of a magnetic resonance imaging (MRI) system suitable for use with the disclosed techniques is shown;
[0012] Figure 2 shows a schematic diagram depicting noise characteristics accompanying SWI multi-echo processing according to aspects of the present disclosure;
[0013] Figure 3 shows a noise histogram representing the noise of the amplitude and phase of the obtained combined echo (the difference between the echo based on the clean data and the echo with the noise added) according to aspects of the present disclosure;
[0014] Figure 4 QQ plot showing the noise obtained in an imaginary channel (difference between echoes based on clean data and data with echoes added with noise) relative to an imaginary channel with echoes added with noise according to aspects of the present disclosure;
[0015] Figure 5 A noise histogram showing the noise obtained by combining the real part of the echo and the imaginary part of the echo (the difference between the echo based on the clean data and the echo with the noise added) according to aspects of the present disclosure;
[0016] Fig. 6A and Figure 6B A flow chart showing a method for improving image quality of SWI (eg, generating a denoised phase image and a denoised magnitude image from a complex image) according to aspects of the present disclosure;
[0017] Fig. 7A and Figure 7B A flow chart showing a method for improving the image quality of SWI (eg, generating only a noise-reduced phase image from a complex image) according to aspects of the present disclosure;
[0018] Figure 8 shows minimum intensity projection images derived from SWI images generated with or without the disclosed technique;
[0019] Fig. 9 SWI images and phase images generated with and without the disclosed technique are shown;
[0020] Fig.10 shows phase images and SWI images acquired with or without noise reduction; and
[0021] Fig.11 Three-dimensional (3D) images acquired using a susceptibility-weighted angiography (SWAN) sequence with or without different SWI processing are shown. DETAILED DESCRIPTION
[0022] 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.
[0023] 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.
[0024] Although various aspects of the following discussion are provided in the context of medical imaging, it should be understood that the technology disclosed in the present invention 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 realistic specific implementations and applications. However, the technology disclosed in the present invention 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 technology disclosed in the present invention can be used in any imaging or screening context or image processing or photography field, in which a group or class of collected data undergoes a reconstruction process to generate an image or volume.
[0025] The deep learning (DL) methods discussed herein may be based on artificial neural networks and may therefore include 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, 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.
[0026] 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 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.
[0027] In MRI, the noise in the acquired complex k-space and in the complex images is Gaussian. However, the noise in the magnitude and phase images is no longer Gaussian. Denoising algorithms have been shown to be better at reducing Gaussian noise and therefore may not be suitable for magnitude and phase images. To address the Gaussian form of noise, techniques have been used to reduce the noise of the real and imaginary channels separately.
[0028] In addition, it has been found that using a multi-component approach to a complex for noise reduction is superior to other techniques, but this works on a single echo. In the case of multiple echoes, noise reduction can be performed on a single echo. However, through acceleration and derating, the signal-to-noise ratio (SNR) decreases and the SNR of the echo is then very poor. Denoising these echoes would mean that the model would have to work aggressively, resulting in blurring. The degree of blurring and any adverse effects caused by noise reduction (e.g., half-pixel shifts) will vary from echo to echo, and combining them after noise reduction will result in artifacts and excessive blurring.
[0029] The present disclosure provides a system and method for improving the image quality of susceptibility weighted imaging (SWI) using deep learning. Specifically, the present disclosure provides a SWI processing pipeline that utilizes denoising (through a trained deep learning-based denoising network or model) in a complex domain (with both real and imaginary data), and then utilizes super-resolution to overcome (e.g., remove) any blurring present in the denoised complex image that may be caused by the denoising model to present a higher resolution complex image (e.g., presenting clear blood vessels).
[0030] The disclosed system and method include acquiring an MRI complex signal having multiple complex echoes of a region of interest (e.g., brain) from a single channel (e.g., element or loop) of a coil (e.g., RF receive coil) by an MRI scanner during a SWI sequence (e.g., a gradient echo (GRE) sequence, such as a three-dimensional (3D) GRE sequence). It should be noted that the disclosed system and method can be used on MRI signals acquired from multiple channels of the coil during the SWI sequence. The disclosed system and method also include phase filtering each of the multiple complex echoes. The disclosed system and method also include generating a corresponding phase image and a corresponding amplitude image from each phase-filtered complex echo. The disclosed system and method even further include combining the corresponding amplitude images of the multiple complex echoes with each other (e.g., by root mean square) to generate a combined amplitude image, and combining the corresponding phase images of the multiple complex echoes with each other (e.g., by averaging) to generate a combined phase image. The disclosed system and method further include generating a complex image from both the combined amplitude image and the combined phase image. The disclosed systems and methods also include denoising the complex image using a deep learning-based denoising network to generate a denoised complex image. The disclosed systems and methods also include applying super-resolution to the denoised complex image to generate a higher resolution fully denoised complex image without any blur (e.g., relative to the denoised complex image before applying super-resolution).
[0031] In some embodiments, the disclosed systems and methods even include generating a phase image from a higher resolution fully denoised complex image, wherein the phase image is fully denoised. In some embodiments, the phase image may be used only to generate (e.g., by scaling) a phase mask. In some embodiments, the phase mask and the corresponding magnitude image of each complex echo in the plurality of complex echoes are used to generate a SWI processed image of the region of interest. In some embodiments, the disclosed systems and methods also include generating a magnitude image from a higher resolution denoised complex image, wherein the magnitude image is partially denoised. In some embodiments, the magnitude image (with a desired percentage (residual percentage) of the residual noise generated by denoising the undenoised complex image added) and the phase image may be used to generate a phase mask. In this scenario, the phase mask is used in conjunction with the magnitude image (with the residual percentage) to generate a SWI processed image of the region of interest. In some embodiments, if the SNR is sufficient for the last complex echo, a single complex echo denoising and super-resolution may be used.
[0032] As in SWI, with the disclosed embodiments, the magnitude image is primarily used to identify the location of the lesion. The denoised complex image (using both the magnitude image (with the residual percentage added) and the phase image or just the phase image derived from the denoised complex image) is used to generate the SWI processed image. The benefit of this approach is that the magnitude image retains its texture and the person reading the image finds the processed image more natural (as opposed to overly cleaned). Since small amounts of noise appear as hypointense spots in the SWI processed image, the phase image is always completely denoised.
[0033] SWI is an important sequence for locating hemorrhage. Value is added when the acquisition of data is accelerated and thin sections are utilized to accurately locate lesions. However, this will drastically reduce the SNR. When this is extended to acquisitions with derated number of coils and field strength, the SNR drops even further to the point of producing non-diagnostic images. The disclosed techniques improve the quality of the images by improving the SNR in these scenarios. The disclosed techniques also reduce artifacts affected by phase noise. The disclosed techniques further render clear blood vessels.
[0034] 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 MRI system 100 is generally configured to perform MR imaging.
[0035] 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 MRI 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.
[0036] 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.
[0037] The various coils of the system 100 are controlled by external circuits 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 amplification and control circuitry for supplying current to the coils as defined by the digitized pulse sequence output by the scanner control circuit 104.
[0038] 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 configured to generate RF pulses. Similarly, the receiving coil 138 is connected to a switch 154, which 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 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.
[0039] It should be noted that although the scanner 102 and control / amplification circuits described above are shown 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 106.
[0040] As shown, the scanner control circuit 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.
[0041] 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.
[0042] The interface circuit 164 is coupled to the control and analysis circuit 160 for exchanging data between the scanner control circuit 104 and the system control circuit 106. In some embodiments, the control and analysis circuit 160, although shown as a single unit, may include one or more hardware devices. The system control circuit 106 includes an interface circuit 166 that receives data from the scanner control circuit 104 and transmits data and commands back to the scanner control circuit 104. The control and analysis circuit 168 may include a CPU in a general or special-purpose computer or workstation. The control and analysis circuit 168 is coupled to a memory circuit 170 to store programming code for operating the MRI 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 the reconstruction of the acquired data as described below when executed by the processor. In some embodiments, the memory circuit 170 may store one or more neural networks for the reconstruction of the acquired data as described below. In some embodiments, image reconstruction may occur on a separate computing device having processing circuits and memory circuits.
[0043] 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.
[0044] Figure 2 A schematic diagram depicting the noise characteristics associated with SWI multi-echo processing is shown (e.g., for Fig. 6A , Figure 6B , Fig. 7A and Figure 7B In order to study the noise characteristics associated with SWI multi-echo processing, some assumptions are made. First, three echoes are collected. In some embodiments, the number of echoes collected can be different. Second, the noise in all real channels and all virtual channels of all echoes is an independent Gaussian distribution, and the SNR decreases for subsequent echoes. Third, the noise characteristics of a single channel are simple and clear. It is assumed that in the case of multiple channels (when calibration-based channel combination is performed), the noise is still Gaussian noise even after the channel combination is performed.
[0045] like Figure 2 As shown, in the Cartesian plane 180, for each echo (e.g., echo 1, echo 2, and echo 3) of the real channel 182 and the imaginary channel 184, the noise distribution is a Gaussian distribution. In the polar plane 186, for the magnitude image 188 of each echo (reconstructed from the real channel 182), the noise distribution becomes a Rician distribution. In the polar plane 186, for the phase image 190 of each echo (generated by phase filtering the imaginary channel 184), the noise distribution becomes a uniform distribution.
[0046] like Figure 2 As shown in , the magnitude images 188 of all echoes are combined using a root mean square 194 (i.e., the square root of the mean square of the magnitude images) to generate a combined magnitude image 196 in the polar plane 86. This enables the collection of structural information from the first echo and the collection of susceptibility information from subsequent echoes. The noise distribution of the combined magnitude image 196 becomes a chi distribution. As shown in Figure 2As shown in FIG. 1 , the phase images 190 of all echoes are combined and averaged 198 to generate a combined phase image 200 in the polar plane. The noise distribution of the combined phase image 196 becomes a triangular distribution. When the combined magnitude image 196 and the combined phase image 200 are combined and then brought back into the Cartesian plane 180, the following function, r*exp(jθ), is applied to generate a complex image, where r is the magnitude, j is the square root of -1, and θ is the angle relative to the real axis.
[0047] Figure 2 The noise distribution in the complex image generated in the simulation is too complex to be calculated. Therefore, simulation is used to study the noise distribution in the complex image. When studying the noise characteristics by simulation, a 3-echo single-channel SWI acquisition is used as the clean data. Independent Gaussian noise is added to the real channel and the imaginary channel, which increases the standard deviation of the subsequent echoes. The echoes of the clean data have the root mean square and average values, respectively. Figure 2 The amplitude image and phase image are combined separately as shown in FIG. In addition, the echo with noise added has the advantages of using the root mean square and averaging respectively. Figure 2 The amplitude image and the phase image are combined separately. Clean data is subtracted from noisy data to obtain noise, which is removed by a noise reduction model (e.g., a trained deep learning-based network or model). Figure 3 Noise histograms 202 and 204 are shown representing the obtained noise (based on the difference between clean data and data with noise added) for the real channel of the combined echo and the imaginary channel of the combined echo, respectively. Figure 3 As shown, in the noise histograms 202 and 204 , the noise distribution is close to a Gaussian distribution, but is definitely not a Gaussian distribution. Figure 4 A QQ plot 206 is shown. The QQ plot 206 plots the noise obtained in the imaginary channel relative to the imaginary channel with the echo added with noise (based on the difference between the clean data and the data with the noise added). The QQ plot 206 shows the fuzziness with the tail. The p-value for a statistical test such as the Shapiro-Wilk test is 1.99*e-23. The QQ plot 206 and the p-value demonstrate that the distribution of the noise obtained in the real and imaginary channels is not Gaussian. Figure 5 Noise histograms 208 and 210 are shown representing the obtained noise (based on the difference between clean data and data with noise added) representing the amplitude of the combined echo and the phase of the combined echo, respectively. The noise distribution in the noise histogram 208 is a Rice distribution. The noise distribution in the noise histogram 210 is a uniform distribution.
[0048] It has been found that a deep learning-based denoising model trained with Gaussian noise simulation and enhanced with a point spread function performs well on complex images derived from a combined magnitude image and a combined phase image of multiple echoes (e.g., Figure 2 It is effective in reducing noise.
[0049] Fig. 6A and Figure 6B A flow chart of a method 212 for improving the image quality of SWI is shown. One or more steps of the method 212 may be performed by a remote computing system or Figure 1 The processing circuit of the magnetic resonance imaging system 100 in the embodiment of the present invention is executed. The processing circuit of the magnetic resonance imaging system 100 in the embodiment of the present invention can be executed simultaneously and / or according to the Fig. 6A and Figure 6B The different orders shown perform one or more steps of method 212. Although described with reference to data acquired from multiple echoes of a single channel of a coil (eg, an RF receive coil), method 212 may be used on data acquired from multiple echoes of multiple channels.
[0050] The method 212 includes scanning an MRI scanner (eg, a 3D GRE sequence) during a SWI sequence (eg, a gradient echo (GRE) sequence, such as a 3D GRE sequence). Figure 1 The MRI scanner 102 in the embodiment of the present invention acquires an MRI complex signal 214 having multiple echoes of a region of interest (e.g., brain) from a single channel (e.g., element or loop) of a coil (e.g., an RF receive coil) (block 216). The method 212 also includes obtaining multiple complex echoes from the MRI signal (block 222).
[0051] The method 212 also includes performing phase filtering on each of the multiple complex echoes (box 223). The method 212 also includes generating (e.g., reconstructing) a corresponding amplitude image 224 and a corresponding phase image 221 from each phase-filtered complex echo (box 226). The method 212 even further includes combining the corresponding amplitude images 224 of the multiple complex echoes with each other to generate a combined amplitude image 228, and combining the corresponding phase images 221 of the multiple complex echoes with each other to generate a combined phase image 230 (box 232). In some embodiments, the combined amplitude image 228 is derived by utilizing the root mean square of the amplitude image 224 (i.e., the square root of the mean square of the amplitude image 224). In some embodiments, the combined phase image 230 is derived by combining and averaging the phase images 221.
[0052] The method 212 further includes generating a complex image 234 (e.g., a complex image that has not been denoised) from both the combined magnitude image 228 and the combined phase image 230 (box 236). In certain embodiments, the combined magnitude image 228 and the combined phase image 230 are combined and then the following function, r*exp(jθ), is applied thereto to generate the complex image 234. The method 212 also includes denoising the complex image 234 using a deep learning-based denoising network or model 237 (e.g., trained with a fully enhanced Gaussian noise simulation) to generate a denoised complex image 238 (box 240). The method 212 also includes applying super-resolution to the denoised complex image 238 to generate a higher resolution fully denoised complex image 242 without any blur (e.g., relative to the denoised complex image 238 before applying super-resolution) (box 244).
[0053] The method 212 also includes generating a phase image 246 (e.g., a filtered phase image) from the higher resolution denoised complex image 242 using phase filtering (box 248). The phase image 246 is always fully denoised. The method 212 also includes generating a magnitude image 250 from the higher resolution denoised complex image 242 (box 252). The magnitude image 250 is partially denoised. In some embodiments, the method 212 includes obtaining a desired percentage of residual noise (residual percentage) 254 by denoising the undenoised complex image 234 to obtain the denoised complex image 238 (box 256). The method 212 also includes adding the residual percentage 254 to the magnitude image 250 to generate the magnitude image 250 (e.g., with the residual percentage 254) (box 258). The residual percentage 254 provides texture. The method 212 also includes using both the magnitude image 259 and the phase image 246 to which the residual percentage is added to generate a phase mask 260 (box 262). The magnitude information of the magnitude image 259 (with the residual percentage added) is used to suppress the background when generating the phase mask 260. The method 212 even includes generating a SWI-processed image 264 of the region of interest using the phase mask 260 and the magnitude image 250 (with the residual percentage 254 added) (block 266). Specifically, the magnitude image 250 (with the residual percentage 254 added) is multiplied by the phase mask 260 a specific number of times.
[0054] Fig. 7A and Figure 7B A flow chart of a method 268 for improving the image quality of SWI is shown. One or more steps of the method 268 may be performed by a remote computing system or Figure 1 The processing circuit of the magnetic resonance imaging system 100 in the embodiment of the present invention is executed. The processing circuit of the magnetic resonance imaging system 100 in the embodiment of the present invention can be executed simultaneously and / or according to the Fig. 7A and Figure 7BThe different orders shown perform one or more steps of method 268. Although described with respect to data acquired from multiple echoes of a single channel of a coil (eg, an RF receive coil), method 268 may be used with data acquired from multiple echoes of multiple channels.
[0055] The method 268 includes obtaining a MRI image during a SWI sequence (eg, a gradient echo (GRE) sequence, such as a three-dimensional (3D) GRE sequence) by a MRI scanner (eg, Figure 1 The MRI scanner 102 in the embodiment of the present invention acquires an MRI complex signal 214 having multiple echoes of a region of interest (e.g., brain) from a single channel (e.g., element or loop) of a coil (e.g., an RF receive coil) (block 270). The method 268 also includes obtaining multiple complex echoes from the MRI signal (block 272).
[0056] The method 268 also includes phase filtering each of the multiple complex echoes (box 274). The method 268 also includes generating (e.g., reconstructing) a corresponding magnitude image 224 and a corresponding phase image 221 from each phase-filtered complex echo (box 276). The method 268 even further includes combining the corresponding magnitude images 224 of the multiple complex echoes with each other to generate a combined magnitude image 228, and combining the corresponding phase images 221 of the multiple complex echoes with each other to generate a combined phase image 230 (box 278). In some embodiments, the combined magnitude image 228 is derived by utilizing the root mean square of the magnitude image 224 (i.e., the square root of the mean square of the magnitude image 224). In some embodiments, the combined phase image 230 is derived by combining and averaging the phase images 220.
[0057] The method 268 further includes generating a complex image 234 (e.g., a complex image that has not been denoised) from both the combined magnitude image 228 and the combined phase image 230 (box 280). In certain embodiments, the combined magnitude image 228 and the combined phase image 230 are combined and then the following function, r*exp(jθ), is applied thereto to generate the complex image 234. The method 268 also includes denoising the complex image 234 using a deep learning-based denoising network or model 237 (e.g., trained with a fully enhanced Gaussian noise simulation) to generate a denoised complex image 238 (box 282). The method 268 also includes applying super-resolution to the denoised complex image 238 to generate a higher resolution fully denoised complex image 242 without any blur (e.g., relative to the denoised complex image 238 before applying super-resolution) (box 284).
[0058] The method 268 also includes generating a phase image 246 (e.g., a filtered phase image) from the higher resolution denoised complex image 242 using phase filtering (box 286). The phase image 246 is always fully denoised. The method 268 also includes generating a phase mask 260 using only the phase image 246 (box 288). The method 268 even includes generating a SWI-processed image 264 of the region of interest using the corresponding magnitude image 224 and the phase mask 260 for each of the plurality of echoes (box 290). Specifically, the corresponding magnitude image 224 is multiplied by the phase mask 260 a specific number of times.
[0059] Figure 8 It shows the difference between using and not using the disclosed technology (e.g., Fig. 6A and Figure 6B The minimum intensity projection image is derived from the SWI image generated by method 212 in . The images were acquired using a 1.5 Tesla (T) MRI scanner using a head and neck unit coil. The acquisition was a 512×384 acquisition with an array coil spatial sensitivity encoding (ASSET) acceleration of 2. Similarly, the acquisition had 5 echoes, a flip angle of 20 degrees, an echo time (TE) of 46.6 milliseconds (ms), a slice thickness of 2 millimeters (mm), 32 positions (slices) per slab, and a total acquisition time of 5 minutes.
[0060] Images 292 and 294 are minimum intensity projection images (e.g., of the brain) obtained from typical SWI-processed images. Images 296 and 298 are images obtained from images processed using Fig. 6A and Figure 6B A minimum intensity projection image (e.g., of the brain) obtained from a SWI-processed image processed by the method 212 in . Specifically, noise reduction is applied after combining the echoes (i.e., noise reduction is performed on a complex image generated from both a combined magnitude image and a combined phase image of multiple echoes). The noise reduction result is only used to generate a phase map. Specifically, the noise-reduced phase image generated from the complex image is only used to generate a phase mask, and the phase mask and magnitude image derived from the echoes are used to generate a SWI-processed image, which is used to generate minimum projection images 296 and 298. Minimum projection images 296 and 298 show that the disclosed technique does not affect clarity, which is important in SWI.
[0061] Fig. 9 The disclosed techniques (e.g., Fig. 6A and Figure 6BSWI images and phase images generated by method 212 in . Images were acquired using a 1.5 Tesla (T) MRI scanner using a head and neck unit coil. The acquisition was a 512×512 acquisition with an array coil spatial sensitivity encoding (ASSET) acceleration of 2. Again, the acquisition had 5 echoes, a flip angle of 20 degrees, an echo time (TE) of 46.6 ms, a slice thickness of 2 mm, 24 positions (slices) per slab, and a total acquisition time of 4 minutes and 30 seconds.
[0062] Image 300 is a SWI image derived by a typical SWI process. Image 302 is a phase image used to obtain SWI image 300. Image 304 is a phase image obtained by using Fig. 6A and Figure 6B 304 . Specifically, noise reduction is applied after combining the echoes (i.e., noise reduction is performed on the complex image generated from both the combined amplitude image and the combined phase image of the plurality of echoes). Image 306 is a phase image (a phase image after noise reduction and super-resolution) used to obtain the SWI image 304. Image 308 is a scaled cortex of image 302. Image 310 is a scaled cortex of image 306. Images 308 and 310 enable better observation of gray-white matter differentiation. Comparison of the images shows that the SNR of the SWI processed images using the disclosed technology is greater (i.e., the SNR of images 304 and 306 is greater than that of 302 and 304). In addition, gray-white matter differentiation in the cortical region in phase image 310 is better than that in phase image 308.
[0063] Fig.10 Phase images and SWI images acquired with or without noise reduction are shown. The images were acquired using a 1.5 Tesla (T) MRI scanner using a body coil. The acquisition was a 356×206 acquisition. Again, the acquisition had 3 echoes, a flip angle of 20 degrees, an echo time (TE) of 34.1 ms, a repetition time (TR) of 49.6 ms, and 38 positions (slices) per slab. Complex images for these images were generated from the combined magnitude images and combined phase images of the multiple echoes as described above.
[0064] Image 312 is a phase image derived from a complex image without noise reduction (and super resolution). Image 314 is a SWI image derived from both a magnitude image and a phase mask (derived from the phase image) derived from a complex image without any noise reduction. Image 316 is a phase image derived from a complex image with full (100%) noise reduction and super resolution applied. Image 318 is a SWI image derived from both a magnitude image of the echo (without noise reduction) and a phase mask (derived from the phase image derived from the complex image after noise reduction and super resolution). Image 320 is a SWI image derived from both a magnitude image and a phase mask (derived from the phase image) derived from a complex image after noise reduction and super resolution. The magnitude image used to derive image 320 was subjected to 65% noise reduction. Image 322 is a SWI image derived from both the magnitude image derived from the complex image after noise reduction and super resolution and the phase mask (derived from the phase image). The magnitude image used to derive image 322 is subjected to 55% noise reduction. The SNR and clarity of image 316 are improved compared to image 312. Similarly, the SNR and clarity of images 318, 320, and 322 are improved to varying degrees compared to image 314.
[0065] Fig.11 3D images acquired using a Susceptibility Weighted Angiography (SWAN) sequence with or without different SWI processing are shown. The images were acquired using a 0.5 Tesla (T) MRI scanner using a head and neck unit coil. The acquisition was a 256×136 acquisition using a SWAN sequence (i.e., a high resolution 3D multi-echo gradient echo sequence). Again, the echoes for this acquisition were 3 echoes, a flip angle of 30 degrees, an echo time (TE) of 90 ms, a repetition time (TR) of 109.5 ms, and 34 positions (slices) per slab.
[0066] Image 324 is a 3D SWAN image that has not been subjected to SWI processing. Image 326 is a 3D SWAN image that has been subjected to SWI processing but has not been subjected to noise reduction. Specifically, a complex image for image 326 is generated from both the combined magnitude image and the combined phase image of the plurality of echoes as described above. Image 328 is a 3D SWAN image that has been subjected to SWI processing and noise reduction (and super-resolution) as described above. Specifically, a complex image for image 328 is generated from both the combined magnitude image and the combined phase image of the plurality of echoes and subjected to noise reduction (and super-resolution).
[0067] It should be noted that the above techniques can be used with any multi-echo acquisition sequence (even amplitude-dependent sequences). For example, the above techniques can be used on scan data acquired using Multi-Echo Reconstructed Gradient Echo (MERGE), which is a destructive T2*-weighted sequence used for spinal and musculoskeletal imaging.
[0068] Technical effects of the disclosed subject matter include providing a SWI processing pipeline that utilizes denoising (via a trained deep learning-based denoising network or model) in the complex domain (with both real and imaginary data), followed by super-resolution to overcome (e.g., remove) any blurring present in the denoised complex image that may be caused by the denoising model to render a higher resolution complex image (e.g., render clear blood vessels). Technical effects of the disclosed subject matter also include improving the quality of the image by improving the SNR. Technical effects of the disclosed subject matter also include reducing artifacts affected by phase noise. Technical effects of the disclosed subject matter even include rendering clear blood vessels.
[0069] Reference is made to the technology presented and claimed herein 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).
[0070] 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 computer-implemented method for improving the image quality of susceptibility-weighted imaging (SWI), comprising: Acquiring, by a processor, a magnetic resonance imaging (MRI) complex signal (214) having a plurality of complex echoes of a region of interest from a coil by an MRI scanner (102) during a SWI sequence; performing phase filtering on each of the plurality of complex echoes by the processor; Generating, by the processor, a corresponding phase image (221) and a corresponding magnitude image (224) from each phase-filtered complex echo; The processor combines the corresponding magnitude images (224) of the plurality of complex echoes with each other to generate a combined magnitude image (228), and combines the corresponding phase images (221) of the plurality of complex echoes with each other to generate a combined phase image (230); generating, by the processor, a complex image (234) from both the combined magnitude image (228) and the combined phase image (230); as well as The processor performs denoising on the complex image (234) using a deep learning-based denoising network (237) to generate a denoised complex image (238).
2. The computer-implemented method of claim 1, further comprising applying, by the processor, super-resolution to the denoised complex image (238) to generate a higher resolution fully denoised complex image (242) without any blur.
3. The computer-implemented method of claim 2, further comprising generating, by the processor, a phase image (246) from the higher resolution fully denoised complex image (242) using phase filtering.
4. The computer-implemented method of claim 3, further comprising generating, by the processor, a phase mask (260) based solely on the phase image (246).
5. The computer-implemented method of claim 4, further comprising generating, by the processor, a SWI-processed image (264) of the region of interest based on the phase mask (260) and the corresponding magnitude image (224) of each of the plurality of complex echoes.
6. The computer-implemented method of claim 3, further comprising generating, by the processor, a magnitude image (250) from the higher resolution fully denoised complex image (242), wherein: The magnitude image is partially denoised (250).
7. The computer-implemented method of claim 6, further comprising generating, by the processor, a phase mask (260) based on a phase image (246) and the magnitude image (250) with a residual percentage (254) of the combined magnitude image (228) added thereto.
8. The computer-implemented method of claim 7, further comprising generating, by the processor, a SWI-processed image (264) of the region of interest based on the phase mask (260) and the magnitude image (250) to which the residual percentage (254) of the combined magnitude image (228) is added.
9. The computer-implemented method of claim 1 , wherein: Combining the corresponding magnitude images (224) of the plurality of complex echoes with one another includes combining the corresponding magnitude images (224) with one another using a root mean square to generate the combined magnitude image (228).
10. The computer-implemented method of claim 1, wherein: Combining the corresponding phase images (221) of the plurality of complex echoes with one another includes combining the corresponding phase images (221) with one another using averaging to generate the combined phase image (230).
11. A system for improving the image quality of susceptibility weighted imaging (SWI), 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: Acquiring a magnetic resonance imaging (MRI) complex signal (214) having a plurality of complex echoes of a region of interest from a coil by an MRI scanner (102) during a SWI sequence; performing phase filtering on each of the multiple complex echoes; Generating a corresponding phase image (221) and a corresponding magnitude image (224) from each phase-filtered complex echo; Respectively combining the corresponding magnitude images (224) of the plurality of complex echoes with one another to generate a combined magnitude image (228), and combining the corresponding phase images (221) of the plurality of complex echoes with one another to generate a combined phase image (230); From the combined magnitude image (228) and the combined phase image (230) Both generate a complex image (234); and The complex image (234) is denoised using a deep learning-based denoising network (237) to generate a denoised complex image (238).
12. The system according to claim 11, wherein: The routine, when executed by the processor, causes the processor to apply super-resolution to the denoised complex image (238) to generate a higher resolution fully denoised complex image (242) without any blur.
13. The system according to claim 12, wherein: The routine, when executed by the processor, causes the processor to generate a phase image (246) from the higher resolution fully denoised complex image (242).
14. The system according to claim 13, wherein: The routine, when executed by the processor, causes the processor to generate a phase mask (260) based solely on the phase image (242).
15. The system of claim 14, wherein: The routine, when executed by the processor, causes the processor to generate a SWI-processed image (264) of the region of interest based on the phase mask (260) and the corresponding magnitude image (224) of each complex echo of the plurality of complex echoes.
