CROSS-REFERENCE TO RELATED APPLICATION(S)

[0001] This application relates to and claims priority from U.S. Patent Application No. 63/125,745, filed on December 15, 2020, the entire disclosure of which is incorporated herein by reference.

FIELD OF THE DISCLOSURE

[0002] The present disclosure relates generally to magnetic resonance imaging (“MRI”), and more specifically, to exemplary embodiments of an exemplary system, method, and computer-accessible medium for facilitating multi-contrast denoising of one or more magnetic resonance images using a native noise structure.

BACKGROUND INFORMATION

[0003] The benefits of deep learning (“DL”) based denoising of MR images can include reduced acquisition time (see, e.g, Reference 1) and improved image quality at low field strength. (See, e.g., Reference 2). However, most studies involve simulating the noise level relative to the signal and its structure, for training the models. These simulations can require biophysical models that can incorporate a variety of tissue parameters that are field and acquisition dependent. (See, e.g., Reference 2). Scaling these simulations may be complex and computationally intensive. Additionally, this can require a vast amount of data to train and validate.

[0004] Thus, it may be beneficial to provide exemplary system, method, and computer- accessible medium for facilitating multi-contrast denoising of one or more magnetic resonance images using a native noise structure which can address and/or overcome at least some of the deficiencies described herein above.

SUMMARY OF EXEMPLARY EMBODIMENTS

[0005] An exemplary system, method, and computer-accessible medium for facilitating multi-contrast denoising of one or more magnetic resonance images using a native noise structure can include, for example, extracting noise segments from the dataset by cropping and storing at least one of comers or edges of the at least one first MR image; adding the noise segments to a Human Connectome Project (“HCP”) dataset to generate a noisy HCP dataset, wherein the noise segments are added at a noise level relative to a maximum image intensity level of the MR dataset; training a native noise denoising network (“NNDnef ’) using the HCP dataset and the noisy HCP dataset; and denoising a second MR image using the NNDnet.

[0006] In some exemplary embodiments of the present disclosure, the exemplary system, method, and computer-accessible medium can train the NNDnet using a rectified linear configuration as an activation function.

[0007] In some exemplary embodiments of the present disclosure, the second MR image can include Tl-weighted data. In some exemplary embodiments of the present disclosure, the second MR image can include T2-weighted images from Tailored MR Fingerprinting. In some exemplary embodiments of the present disclosure, the second MR image can include low field brain T1 weighted imaging. In some exemplary embodiments of the present disclosure, the low field brain T1 weighted imaging can include lower signal to noise ratio. [0008] In some exemplary embodiments of the present disclosure, the NNDnet can include a U-net. In some exemplary embodiments of the present disclosure, the NNDnet can be configured to retain edge information. In some exemplary embodiments of the present disclosure, the noise segments can be added at a noise level relative to a maximum image intensity level of the MR dataset.

[0009] These and other objects, features and advantages of the exemplary embodiments of the present disclosure will become apparent upon reading the following detailed description of the exemplary embodiments of the present disclosure, when taken in conjunction with the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Further objects, features and advantages of the present disclosure will become apparent from the following detailed description taken in conjunction with the accompanying Figures showing illustrative embodiments of the present disclosure, in which:

[0011] Figures 1A and IB are exemplary diagrams of an exemplary native noise denoising network according to an exemplary embodiment of the present disclosure;

[0012] Figures 2A-2G are exemplary images generated using a denoising tailored magnetic resonance fingerprinting procedure according to an exemplary embodiment of the present disclosure; [0013] Figures 3A-3G are exemplary T2 images produced using a denoising tailored magnetic resonance fingerprinting according to an exemplary embodiment of the present disclosure;

[0014] Figures 4A-4G are exemplary denoised low field Tl-weighted images according to an exemplary embodiment of the present disclosure;

[0015] Figures 5A-5D are exemplary graphs illustrating image quality evaluation according to an exemplary embodiment of the present disclosure; and

[0016] Figure 6 is an illustration of an exemplary block diagram of an exemplary system in accordance with certain exemplary embodiments of the present disclosure.

[0017] Throughout the drawings, the same reference numerals and characters, unless otherwise stated, are used to denote like features, elements, components or portions of the illustrated embodiments. Moreover, while the present disclosure will now be described in detail with reference to the figures, it is done so in connection with the illustrative embodiments and is not limited by the particular embodiments illustrated in the figure and the appended claims.

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

[0018] The exemplary system, method, and computer-accessible medium, according to an exemplary embodiment of the present disclosure, can utilize the native noise of the data that can be denoised, which can be referred to as a native noise denoising network (“NNDnef ’). The exemplary NNDnet can be applied to three different datatypes: (i) Tl-weighted (ii) T2- weighted images from Tailored MR Fingerprinting (“TMRF”) (see, e.g., References 3 and 4), which can facilitate rapid acquisition of six, non-synthetic contrasts and two quantitative tissue parametric maps; and (iii) low field (e.g., 0.36T) brain T1 weighted imaging, which can suffer from lower signal to noise ratio (“SNR”) compared to the widely used 1.5T system.

Exemplary Method/Procedure

[0019] The exemplary training data which has been tested includes 8295 T1 MPRAGE and 6622 T2 weighted images from the human connectome project. (See, e.g., Reference 5). The forward modeling of noisy data includes extracting noise patches from a target application data set that is noisy. The extraction was performed by cropping and storing the comers of the images. These noisy patches were then collaged and added to Human Connectome Project (“HCP”) data at a noise level relative to the maximum image intensity level found in the native data set. The noisy and clean HCP datasets were used to train the exemplary NNDnet using an exemplary U-net (see Figures 1 A and IB) with the rectified linear unit as the activation function over 400 epochs on a four GPU computer.

[0020] Figures 1A and IB illustrate exemplary diagrams of an exemplary native noise denoising network according to an exemplary embodiment of the present disclosure. In particular, Figure 1A provides an illustration of an exemplary block diagram in which the exemplary embodiments of the method, system and computer-accessible medium according to the present disclosure can utilize an exemplary forward model extracts noise patches from the training data (e.g., in procedure 120) and adds them to the human connectome data (e.g., in procedure 130) to train the native noise denoising model (NNDnet) e.g., in procedure 140). Such exemplary noise addition can alleviate the need for a native gold standard data acquisition while retaining the noise structure and levels of the target application. Figure IB shows an exemplary diagram of an exemplary NNDnet’ s neural network architecture 400 which can include a U-net using the rectified linear unit activation function according to an exemplary embodiment of the present disclosure.

[0021] The exemplary model was used to determine noise and signal levels. Previously acquired Tl-weighted images on a 0.36T Mindray and TMRF data on a 3T GE Premier were used. For the three types of data, noise patches of size 14 x 14 from 50% of the data were extracted and the other half was tested. This resulted in testing 3420 slices for low field imaging, 2220 slices for TMRF- Tl, and 900 slices for TMRF-T2. The training denoising performance of the exemplary NNDnet images were evaluated using peak SNR (“PSNR”) with respect to the clean HCP data. The test images from the three applications were denoised using a gradient anisotropic diflusion denoising (“AD”) in a 3D Slicer tool (see, e.g., Reference 6), the exemplary NNDnet, and a combination of the two exemplary denoising methods. The test images from the three applications did not have a gold standard (e.g., no reference). Thus, the image entropy that reflects the detail in an image was calculated for the three denoising combinations and compared.

Exemplary Results and Discussion

[0022] The training performance of the exemplary NNDnet is shown in Figures 2A-2C. In particular, Figure 2C illustrates an exemplary denoised image which is similar to the image shown in Figure 2A. The training was performed for twenty -two hours on a four GPU computer. The exemplary noise structure and relative amplitude to the signal shown in Figure 2B is reflected in the test image shown in Figure 2D. The exemplary AD filtering results in blurring shown in Figure 2E, indicates that the exemplary NNDnet retains the edge information. (See e.g., Figure 2F). The combination of the two denoising methods provides a balance between edge preservation and denoising.

[0023] In particular, Figures 2A-2G show exemplary images which are generated using an exemplary denoising tailored MR Fingerprinting (TMRF) procedure according to an exemplary embodiment of the present disclosure. For example, in Figures 2A-2C, the corresponding exemplary magnified images are shown on the left. In Figures 2D-2G, the corresponding exemplary magnified images are shown on the right.

[0024] Further, Figures 2A-2C can illustrate exemplary training images according to an exemplary embodiment of the present disclosure. Indeed, Figure 2A shows an exemplary image of a Tl-weighted image from the human connectome database according to an exemplary embodiment of the present disclosure. Figure 2B illustrates an exemplary image of extracted noise from the TMRF data added for training according to an exemplary embodiment of the present disclosure. Figure 2C shows an exemplary image of corresponding native noise denoising network (NNDnet) result according to an exemplary embodiment of the present disclosure. The left column can show the corresponding magnified images for the red square shown in Figure 2 A.

[0025] Figures 2D-2G illustrate exemplary testing images according to an exemplary embodiment of the present disclosure. Indeed, Figure 2D shows an exemplary image of a test TMRF T1 image that sufiers from noise e) corresponding gradient anisotropy difiusion denoised (GADD) result according to an exemplary embodiment of the present disclosure. Figure 2F shows an exemplary NNDnet denoised image according to an exemplary embodiment of the present disclosure. Figure 2G illustrates an exemplary image of NNDnet + GADD denoised image according to an exemplary embodiment of the present disclosure. [0026] Figures 3A-3G and 4A-4G illustrate similar exemplary representative results for TMRF T2 and low field T1 denoising.

[0027] For example, Figures 3A-3G show exemplary denoising tailored MR Fingerprinting derived T2-weighted images procedure according to an exemplary embodiment of the present disclosure. In particular, in Figures 3A-3C, the exemplary corresponding magnified images are shown on the left. Further, in Figures 3D-3G, the exemplary corresponding magnified images are shown on the right. [0028] Indeed, Figures 3A-3C can provide exemplary training images according to an exemplary embodiment of the present disclosure. For example, Figure 3A illustrates an exemplary T2-weighted label image from the human connectome database according to an exemplary embodiment of the present disclosure. Figure 3B shows an exemplary noise added image which can be used as an input for the training according to an exemplary embodiment of the present disclosure. Figure 3C illustrates an exemplary output of the native noise denoising network (NNDnet) according to an exemplary embodiment of the present disclosure. The corresponding magnified images of the training data are provided on the left for the red (or bright) square shown in Figure 3A.

[0029] Further, Figures 3D-3G can provide exemplary testing images according to an exemplary embodiment of the present disclosure. For example, Figure 3D shows an exemplary TMRF T2 noisy image according to an exemplary embodiment of the present disclosure. Figure 3E illustrates an exemplary gradient anisotropy diffusion denoised (AD) result according to an exemplary embodiment of the present disclosure. Figure 3F shows an exemplary NNDnet denoised image according to an exemplary embodiment of the present disclosure. Figure 3G is an exemplary NNDnet + AD denoised image according to an exemplary embodiment of the present disclosure.

[0030] Figures 4A-4C illustrate representative exemplary reconstructions of one of eleven slices at two TEs. As shown in Figures 4A-4C, the exemplary SER has similar contrast compared to 2D SEGS without any blurring or saturation of intensities.

[0031] For example, Figures 4A-4C can provide exemplary training images according to an exemplary embodiment of the present disclosure. In particular, Figure 4A shows an exemplary image used as the gold standard for training according to an exemplary embodiment of the present disclosure. Figure 4B illustrates an exemplary noise-added-image used for the training according to an exemplary embodiment of the present disclosure. Figure 4C shows an exemplary output of the NNDnet. Indeed, in Figures 4A-4C, the exemplary corresponding magnified images are provided on the left for the red square shown in Figure 4A.

[0032] Further, Figures 4D-4G can provide exemplary testing images according to an exemplary embodiment of the present disclosure. For example, Figure 4D shows an exemplary representative 0.36T noisy image according to an exemplary embodiment of the present disclosure. Figure 4E illustrates an exemplary gradient anisotropy diffusion denoised (AD) result that is blurry according to an exemplary embodiment of the present disclosure. Figure 4F shows an exemplary NNDnet denoised image according to an exemplary embodiment of the present disclosure. Figure 4G illustrates an exemplary NNDnet + AD denoised image according to an exemplary embodiment of the present disclosure. Indeed, in Figures 4D-4G, the right column contains exemplary corresponding magnified images. [0033] Figures 5A-5D shows the exemplary training and testing performance of the exemplary NNDnet. The PSNR for the exemplary NNDnet denoised images increases for the three applications. (See, e.g, Figure 5 A). The mean+Z- SD entropy of AD, NNDnet, and the combination of the two methods computed over slices show NNDnet performing better than AD and the combination of the two methods provides the highest entropy. The denoising of these three dilferent contrasts at two dilferent field strengths demonstrate the benefits of the native noise approach which can include: (i) inherently learning the structure and level of the noise of the specific noisy images, (ii) not requiring the acquisition of gold standard data for the noisy images, and (iii) easily adapting to dilferent noise structures and amplitudes without a vast amount of noisy training data as each image produces four patches of noise to leam. [0034] For example, Figure 5 A illustrates an exemplary graph according to an exemplary embodiment of the present disclosure. Figure 5 A shows exemplary training performance - peak signal to noise ratio (PSNR) of the input and NNDnet denoised images compared to the gold standard training data, for the three applications: low field Tl, tailored MR fingerprinting Tl and T2 imaging. Figure 5B illustrates an exemplary graph according to an exemplary embodiment of the present disclosure providing an exemplary entropy of the input noisy image, gradient anisotropy diffusion denoised (AD) image, the NNDnet denoised image, and NNDnet + AD denoised image for the TMRF Tl weighted images. Figure 5C shows an exemplary graph showing the corresponding gradient entropy measures for the TMRF T2 according to an exemplary embodiment of the present disclosure. Figure 5D illustrates an exemplary graph for the low field Tl weighted images according to an exemplary embodiment of the present disclosure.

[0035] Figure 6 shows a block diagram of an exemplary embodiment of a system according to the present disclosure. For example, exemplary procedures in accordance with the present disclosure described herein can be performed by a processing arrangement and/or a computing arrangement (e.g., computer hardware arrangement) 605. Such processing/computing arrangement 605 can be, for example entirely or a part of, or include, but not limited to, a computer/processor 610 that can include, for example one or more microprocessors, and use instructions stored on a computer-accessible medium (e.g., RAM, ROM, hard drive, or other storage device).

[0036] As shown in Figure 6, for example a computer-accessible medium 615 (e.g., as described herein above, a storage device such as a hard disk, floppy disk, memory stick, CD- ROM, RAM, ROM, etc., or a collection thereol) can be provided (e.g., in communication with the processing arrangement 605). The computer-accessible medium 615 can contain executable instructions 620 thereon. In addition or alternatively, a storage arrangement 625 can be provided separately from the computer-accessible medium 615, which can provide the instructions to the processing arrangement 605 so as to configure the processing arrangement to execute certain exemplary procedures, processes, and methods, as described herein above, for example.

[0037] Further, the exemplary processing arrangement 605 can be provided with or include an input/output ports 635, which can include, for example a wired network, a wireless network, the internet, an intranet, a data collection probe, a sensor, etc. As shown in Figure 6, the exemplary processing arrangement 605 can be in communication with an exemplary display arrangement 630, which, according to certain exemplary embodiments of the present disclosure, can be a touch-screen configured for inputting information to the processing arrangement in addition to outputting information from the processing arrangement, for example. Further, the exemplary display arrangement 630 and/or a storage arrangement 625 can be used to display and/or store data in a user-accessible format and/or user-readable format.

[0038] The foregoing merely illustrates the principles of the disclosure. Various modifications and alterations to the described embodiments will be apparent to those skilled in the art in view of the teachings herein. It will thus be appreciated that those skilled in the art will be able to devise numerous systems, arrangements, and procedures which, although not explicitly shown or described herein, embody the principles of the disclosure and can be thus within the spirit and scope of the disclosure. Various different exemplary embodiments can be used together with one another, as well as interchangeably therewith, as should be understood by those having ordinary skill in the art. In addition, certain terms used in the present disclosure, including the specification, drawings and claims thereof, can be used synonymously in certain instances, including, but not limited to, for example, data and information. It should be understood that, while these words, and/or other words that can be synonymous to one another, can be used synonymously herein, that there can be instances when such words can be intended to not be used synonymously. Further, to the extent that the prior art knowledge has not been explicitly incorporated by reference herein above, it is explicitly incorporated herein in its entirety. All publications referenced are incorporated herein by reference in their entireties.

EXEMPLARY REFERENCES

[0039] The following references are hereby incorporated by reference in their entireties.

1. Keerthi Sravan Ravi, Sairam Geethanath, Patrick Quarterman, Maggie Fung, and John Thomas Vaughan Jr., Intelligent Protocolling for Autonomous MRI, ISMRM 2020.

2. Figini, M., Lin, H., Ogbole, G., Arco, F.D., Blumberg, S.B., Carmichael, D.W., Tanno, R., Kaden, E., Brown, B.J., Lagunju, I. and Cross, H.J., 2020. Image Quality Transfer Enhances Contrast and Resolution of Low-Field Brain MRI in African Paediatric Epilepsy Patients. arXiv preprint arXiv:2003.07216.

3. Sairam Geethanath*, Sachin Jambawalikar, Maggie Fung, Angela Lingelli, John Thomas Vaughan Jr., Rapid, simultaneous non-synthetic multi-contrast andquantitative imaging using Tailored MR Fingerprinting, ISMRM 2019.

4. Pavan Poojar, Enlin Qian, Maggie Fung, and Sairam Geethanath; Natural, multicontrast and quantitative imaging of the brain using tailored MR fingerprinting, ISMRM 2020.

5. Van Essen, D.C., Smith, S.M., Barch, D.M., Behrens, T.E., Yacoub, E., Ugurbil, K. and Wu-Minn HCP Consortium, 2013. The WU-Minn human connectome project: an overview. Neuroimage, 80, pp.62-79.

6. Pieper, S., Lorensen, B., Schroeder, W. and Kikinis, R., 2006, April. The NA-MIC Kit: ITK, VTK, pipelines, grids and 3D slicer as an open platform for the medical image computing community. In 3rd IEEE International Symposium on Biomedical Imaging: Nano to Macro, 2006. (pp. 698-701). IEEE.