DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on October 04, 2024, is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the Examiner.
Claim Objections
Claims 1 and 7 are objected to for minor informalities.
Regarding claim 1, the Examiner recommends amending “wherein the r one or more processors include” to read “wherein the
Regarding claim 7, “wherein the CNN includes a plurality of CNN” should be amended to “wherein the CNN includes a plurality of CNNs”.
Appropriate correction is required.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitations are: “Imaging Unit” in claim 1.
Because this claim limitation is being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it is being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof. This corresponding structure of the imaging unit is a traditional MRI apparatus as described in [0018] of the specification and labeled as 10 in Fig. 1.
If applicant does not intend to have this limitations interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitations to avoid it being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation recites sufficient structure to perform the claimed function so as to avoid it being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-3 and 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (MRI Gibbs-ringing artifact reduction by means of machine learning using convolutional neural networks. Magn Reson Med. 2019; 82: 2133–2145.), hereafter Zhang, in view of Wuelker et al. (EP 4318015 A1), hereafter Wuelker.
Regarding claim 1, Zhang teaches a magnetic resonance imaging apparatus comprising: an imaging unit that collects measurement data consisting of nuclear magnetic resonance signals (For training and testing the CNN, Zhang utilizes images which have been obtained through MRI scanning and reconstruction. [Section 2.2.1] “These images were acquired on a 3.0T scanner (a customized Siemens 3.0T “Connectome Skyra” housed at Washington University in St. Louis, using a standard 32-channel Siemens receiver head coil and a “body” transmission coil designed by Siemens specifically for the small space available using the special gradients of the WU-Minn and MGH-UCLA Connectome scanners.) with a 0.7 mm isotropic resolution.”);
one or more processors that have a function of reconstructing the measurement data at a desired reconstruction matrix size (Section 2.2.1 discusses the training and testing datasets acquired from MRI scanning. Zhang disusses the specific FOV ranges and the pixel size of each scanner used to collect each dataset; thus, images are obtained using a specific reconstruction matrix size. Furthermore, Zhang mentions one dataset being obtained through MRI scanning and IRIS reconstruction, which would require one or more processors to perform the reconstruction. [Section 2.2.1] “The image was reconstructed using the IRIS method.”), and
correcting ringing that occurs in a reconstructed image (Fig. 1 shows the structure of the CNN used for correcting Gibbs ringing artifacts. The CNN receives an MRI image with ringing and outputs a map of the ringing artifacts. Then, the artifact-free image is obtained by subtracting the CNN-estimated artifact map from the input image [Section 2].) in a case in which a measurement matrix size of the measurement data and the reconstruction matrix size are different from each other ([Section 1] “Gibbs-ringing artifact, also known as truncation or spectral leakage artifact, refers to a series of intensity oscillations near sharp edges in MR images. This artifact is caused by insufficient sampling of high-frequency data in the k-space domain and usually appears when the acquisition window limits data acquisition.”),
wherein the r one or more processors include a plurality of CNNs that have been trained by using, as training data, a correct answer image in which ringing has not occurred and an input image in which ringing has occurred ([Section 2.2.1] “The training data consist of 17532 T2-weighted (T2W) brain images of 136 healthy adult subjects from the human connectome project… Axial images were extracted from the 3D data with an original size of 320 × 320 and then resized to 256 × 256 images by cropping and zero-padding. Axial images without anatomical structures were excluded. In total, 17,532 T2W brain images were obtained as the artifact-free images for training. The corresponding images with ringing artifact were obtained by setting a certain percentage of high-frequency components to zero in the k-space domain.”).
As shown above, Zhang teaches ringing correction in MRI images using a CNN; however, Zhang does not teach utilizing several different CNNs for handling different sampling patterns. Specifically, Zhang fails to teach wherein the r one or more processors include a plurality of CNNs to correct the ringing of the input image and that have different sampling patterns of k-space data of the input image, and Zhang fails to teach selecting any one of the plurality of CNNs in accordance with a sampling pattern of the measurement data and apply the selected CNN.
However, Wuelker teaches correct the ringing of the input image and that have different sampling patterns of k-space data of the input image, and select any one of the plurality of CNNs in accordance with a sampling pattern of the measurement data and apply the selected CNN (Wuelker teaches using CNNs for MRI image denoising. [0008] “The different architectures of neural networks which could be used for the group of noise filtering neural networks includes a pulse-coupled neural network, a convolutional neural network, a wavelet neural network, a convolutional neural network-based auto-encoder architecture such as the RED-NET, a UNET, or a ResNet neural network.” Furthermore, Wuelker teaches that CNNs for denoising MRI images may not generalize well across noise power spectrums (NPS) and MRI acquisition schemes, so differently trained CNNs may be used for different denoising problems which are specific to a particular k-space sampling pattern and reconstruction algorithm. [0010] “In another embodiment each of the group of noise filtering neural networks has a specified k-space sampling pattern and a specified reconstruction algorithm. The image metadata further comprises a k-space sampling pattern identifier and a reconstruction algorithm identifier. The selection of the chosen noise filter is at least partially performed by matching the k-space sampling pattern identifier and the reconstruction algorithm identifier to the specified k-space sampling pattern and the specified reconstruction algorithm of each of the group of noise filtering neural networks.” [0065] “However, it has become apparent that a denoising neural network that was trained for a particular noise power spectrum (NPS) will poorly (not at all) generalize to other types of colored noise. Therefore, examples may provide for training different networks (122 group of noise filtering neural networks) for different types of MR acquisition schemes. These are combined into an AI denoising framework with NPS awareness (using the image noise profile 128 to select the chosen noise filtering neural network 130).”).
Zhang and Wuelker are analogous in the art to the claimed invention, because both teach methods of correcting noise in MRI images using one or more neural networks. Therefore, it would have been obvious to one of ordinary skill in the art to modify Zhang’s invention by utilizing different denoising neural networks to handle different sampling pattern types. This modification would improve upon Zhang’s method, since having neural networks trained to handle specific sampling types may improve the denoising quality and performance over using a single network trained on multiple sampling types ([Wuelker 0065] “Described below (in Fig. 6) is an example that uses a radial k-space sampling trajectory that shows a denoising network trained with uncorrelated Gaussian noise will not denoise the image, while a dedicatedly trained network (the chosen noise filtering neural network 130) delivers satisfactory denoising performance.”).
Regarding claim 2, Zhang and Wuelker teach the magnetic resonance imaging apparatus according to claim 1. Zhang further teaches wherein the CNNs have been trained by using, as the input image, an image obtained by cutting out a low-frequency region with a predetermined sampling pattern from k-space data of the correct answer image, performing zero-filling of a high-frequency region, and performing transformation into image data ([Section 2.2.1] “The training data consist of 17532 T2-weighted (T2W) brain images of 136 healthy adult subjects from the human connectome project… Axial images were extracted from the 3D data with an original size of 320 × 320 and then resized to 256 × 256 images by cropping and zero-padding. Axial images without anatomical structures were excluded. In total, 17,532 T2W brain images were obtained as the artifact-free images for training. The corresponding images with ringing artifact were obtained by setting a certain percentage of high-frequency components to zero in the k-space domain.”).
Regarding claim 3, Zhang and Wuelker teach the magnetic resonance imaging apparatus according to claim 1. Wuelker further teaches wherein the one or more processors include, as the plurality of CNNs (Wuelker teaches using different neural networks specific to the noise power spectrum and the MR acquisition scheme. [0010] “In another embodiment each of the group of noise filtering neural networks has a specified k-space sampling pattern and a specified reconstruction algorithm. The image metadata further comprises a k-space sampling pattern identifier and a reconstruction algorithm identifier. The selection of the chosen noise filter is at least partially performed by matching the k-space sampling pattern identifier and the reconstruction algorithm identifier to the specified k-space sampling pattern and the specified reconstruction algorithm of each of the group of noise filtering neural networks.”),
a CNN that has been trained to perform ringing correction on measurement data having a rectangular sampling pattern and a CNN that has been trained to perform ringing correction on measurement data having an elliptical sampling pattern (When discussing obtaining training data for the CNNs, Wuelker mentions that training data may include ringing and artifacts specific to the sampling pattern [0074] “2. In the previous description also the noise free image was converted to k-space and back, i.e. the training target image will have ringing and sub-sampling artifacts associated with the acquisition pattern.” Both rectangular and elliptical sampling patterns are well-known and widely implemented sampling patterns in MRI.).
Therefore, it would have been obvious to one of ordinary skill in the art to modify Zhang’s invention by utilizing different denoising neural networks to handle different sampling pattern types, such as rectangular and elliptical. This modification would improve upon Zhang’s method, since having neural networks trained to handle specific sampling types will improve the denoising quality and performance over using a single network trained to handle multiple different sampling pattern types ([Wuelker 0065] “Described below (in Fig. 6) is an example that uses a radial k-space sampling trajectory that shows a denoising network trained with uncorrelated Gaussian noise will not denoise the image, while a dedicatedly trained network (the chosen noise filtering neural network 130) delivers satisfactory denoising performance.”).
Regarding claim 7, Zhang teaches an image processing method of, in a case in which measurement data collected by a magnetic resonance imaging apparatus is reconstructed (For training and testing the CNN, Zhang utilizes images which have been obtained through MRI scanning and reconstruction. [Section 2.2.1] “These images were acquired on a 3.0T scanner (a customized Siemens 3.0T “Connectome Skyra” housed at Washington University in St. Louis, using a standard 32-channel Siemens receiver head coil and a “body” transmission coil designed by Siemens specifically for the small space available using the special gradients of the WU-Minn and MGH-UCLA Connectome scanners.) with a 0.7 mm isotropic resolution.”) at a reconstruction matrix size different from a matrix size of the measurement data ([Section 1] “Gibbs-ringing artifact, also known as truncation or spectral leakage artifact, refers to a series of intensity oscillations near sharp edges in MR images. This artifact is caused by insufficient sampling of high-frequency data in the k-space domain and usually appears when the acquisition window limits data acquisition.”),
performing ringing correction by using a CNN (Fig. 1 shows the structure of the CNN used for predicting ringing artifacts. The CNN receives an MRI image with ringing and outputs a map of the ringing artifacts. Then, the artifact-free image is obtained by subtracting the CNN-estimated artifact map from the input image [Section 2].) that has been trained by using, as training data, a correct answer image in which ringing has not occurred and an input image in which ringing has occurred, to correct the ringing of the input image ([Section 2.2.1] “The training data consist of 17532 T2-weighted (T2W) brain images of 136 healthy adult subjects from the human connectome project… Axial images were extracted from the 3D data with an original size of 320 × 320 and then resized to 256 × 256 images by cropping and zero-padding. Axial images without anatomical structures were excluded. In total, 17,532 T2W brain images were obtained as the artifact-free images for training. The corresponding images with ringing artifact were obtained by setting a certain percentage of high-frequency components to zero in the k-space domain.”).
As shown above, Zhang teaches ringing correction in MRI images using a CNN; however, Zhang does not teach utilizing several different CNNs for handling different sampling patterns. Specifically, Zhang fails to teach wherein the CNN includes a plurality of CNN that have been trained by using, as the input image, image data generated from a plurality of k-space data having different sampling patterns, and the image processing method comprises: selecting a CNN corresponding to a sampling pattern of the measurement data from among the plurality of CNNs and performing the ringing correction on the measurement data in a real space.
However, Wuelker teaches wherein the CNN includes a plurality of CNN that have been trained by using, as the input image, image data generated from a plurality of k-space data having different sampling patterns, and the image processing method comprises: selecting a CNN corresponding to a sampling pattern of the measurement data from among the plurality of CNNs and performing the ringing correction on the measurement data in a real space (Wuelker teaches using CNNs for MRI image denoising. [0008] “The different architectures of neural networks which could be used for the group of noise filtering neural networks includes a pulse-coupled neural network, a convolutional neural network, a wavelet neural network, a convolutional neural network-based auto-encoder architecture such as the RED-NET, a UNET, or a ResNet neural network.” Furthermore, Wuelker teaches that CNNs for denoising MRI images may not generalize well across noise power spectrums (NPS) and MRI acquisition schemes, so differently trained CNNs may be used for different denoising problems which are specific to a particular k-space sampling pattern and reconstruction algorithm. [0010] “In another embodiment each of the group of noise filtering neural networks has a specified k-space sampling pattern and a specified reconstruction algorithm. The image metadata further comprises a k-space sampling pattern identifier and a reconstruction algorithm identifier. The selection of the chosen noise filter is at least partially performed by matching the k-space sampling pattern identifier and the reconstruction algorithm identifier to the specified k-space sampling pattern and the specified reconstruction algorithm of each of the group of noise filtering neural networks.” [0065] “However, it has become apparent that a denoising neural network that was trained for a particular noise power spectrum (NPS) will poorly (not at all) generalize to other types of colored noise. Therefore, examples may provide for training different networks (122 group of noise filtering neural networks) for different types of MR acquisition schemes. These are combined into an AI denoising framework with NPS awareness (using the image noise profile 128 to select the chosen noise filtering neural network 130).”).
Therefore, it would have been obvious to one of ordinary skill in the art to modify Zhang’s invention by utilizing different denoising neural networks to handle different sampling pattern types. This modification would improve upon Zhang’s method, since having neural networks trained to handle specific sampling types may improve the denoising quality and performance over using a single network trained on multiple sampling types ([Wuelker 0065] “Described below (in Fig. 6) is an example that uses a radial k-space sampling trajectory that shows a denoising network trained with uncorrelated Gaussian noise will not denoise the image, while a dedicatedly trained network (the chosen noise filtering neural network 130) delivers satisfactory denoising performance.”).
Regarding claim 8, Zhang and Wuelker teach the image processing method according to claim 7. Zhang further teaches wherein each of the plurality of CNNs has been trained by using, as the input image data, image data obtained by cutting out a low-frequency region with a predetermined sampling pattern from k-space data of the correct answer image, performing zero-filling of a high-frequency region, and then performing transformation into real space data ([Section 2.2.1] “The training data consist of 17532 T2-weighted (T2W) brain images of 136 healthy adult subjects from the human connectome project… Axial images were extracted from the 3D data with an original size of 320 × 320 and then resized to 256 × 256 images by cropping and zero-padding. Axial images without anatomical structures were excluded. In total, 17,532 T2W brain images were obtained as the artifact-free images for training. The corresponding images with ringing artifact were obtained by setting a certain percentage of high-frequency components to zero in the k-space domain.”).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang (MRI Gibbs-ringing artifact reduction by means of machine learning using convolutional neural networks. Magn Reson Med. 2019; 82: 2133–2145.) and Wuelker (EP 4318015 A1), and further in view of Sutton (US 2010/0194389 A1).
Regarding claim 6, Zhang and Wuelker, teach the magnetic resonance imaging apparatus according to claim 1. However, both Zhang and Wuelker discuss using multiple CNNs for ringing correction based on either the undersampling ratio or the sampling pattern. However, both Zhang and Wuelker do not specifically teach selecting a sampling pattern to use based on the type of scanning. Thus Zhang and Wuelker fail to teach wherein, in a case in which the measurement data is three-dimensional data obtained by raster scanning, the one or more processors select a CNN for a rectangular parallelepiped sampling pattern or a CNN for a rectangular sampling pattern.
However, Sutton teaches wherein, in a case in which the measurement data is three-dimensional data obtained by raster scanning, the one or more processors select a CNN for a rectangular parallelepiped sampling pattern or a CNN for a rectangular sampling pattern (In the Background section, Sutton shows that traditional raster scan patterns are widely known as rectilinear or cartesian scans, and it is known in the art that raster scans produce a cartesian sampling pattern. [0006] “Each MR measurement cycle, or pulse sequence, typically samples a portion of k-space along a sampling trajectory characteristic of that pulse sequence. Most pulse sequences sample k-space in a raster scan-like pattern sometimes referred to as a “spin-warp,” a “Fourier,” a “rectilinear,” or a “Cartesian” scan. The spin-warp scan technique employs a variable amplitude phase encoding magnetic field gradient pulse prior to the acquisition of MR spin-echo signals to phase encode spatial information in the direction of this gradient.”).
Zhang, Wuelker, and Sutton are all analogous in the art to the claimed invention, because all teach methods of operating on MRI images with different sampling patterns. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention taught by Zhang and Wuelker to choose a CNN which handles cartesian sampling (rectangular sampling) when selecting a CNN based on the sampling pattern to perform noise correction. This modification would be obvious to apply since raster scans are known to produce cartesian sampling patterns ([0006] “Each MR measurement cycle, or pulse sequence, typically samples a portion of k-space along a sampling trajectory characteristic of that pulse sequence. Most pulse sequences sample k-space in a raster scan-like pattern sometimes referred to as a “spin-warp,” a “Fourier,” a “rectilinear,” or a “Cartesian” scan.).
Allowable Subject Matter
Claims 4-5 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Regarding claim 4, the closest prior art of record, Zhang (MRI Gibbs-ringing artifact reduction by means of machine learning using convolutional neural networks. Magn Reson Med. 2019; 82: 2133–2145.) and Wuelker (EP 4318015 A1) teach the magnetic resonance imaging apparatus according to claim 3.
However, both Zhang and Wuelker fail to teach wherein the CNN that has been trained to perform the ringing correction on the measurement data having the rectangular sampling pattern and the CNN that has been trained to perform the ringing correction on the measurement data having the elliptical sampling pattern each include a CNN that has been trained to correct ringing in a one-dimensional direction and a CNN that has been trained to correct ringing in a two-dimensional direction.
As discussed in the rejections to claims 1-3, Zhang teaches utilizing multiple CNNs for performing ringing correction in MRI images, but Zhang teaches using different CNNs based on different under-sampling ratios and not based on different sampling patterns. Wuelker teaches utilizing CNNs for removing noise and artifacts (which can include ringing) from MRI images, and Wuelker teaches using different CNNs specifically trained to handle different sampling patterns. However, Wuelker teaches noise correction of many noise types and does not provide specific details about ringing correction regarding specific sampling patterns. Thus, Wuelker does not teach or motivate training additional CNNs for each sampling pattern for correcting ringing in a one-dimensional or a two-dimensional direction.
Other analogous prior art teaches utilizing one or more neural networks and adjusting different machine learning parameters based on the different sampling patterns, but these references fail to apply CNNs to ringing correction, specifically. Furthermore, they do not mention applying a specific CNN to each direction of ringing correction for each sampling pattern. Rather, they teach using one or more neural networks for MRI image reconstruction. For example, Takeshima (US 2019/0362472 A1) teaches activating different layers and parameters of a learned model based on aspects (such as sampling pattern) of the MRI data being reconstructed. Although this is similar to using separate CNNs based on the sampling pattern, Takeshima does not specifically teach using separate CNNs. Furthermore, Takeshima does not teach CNNs for correcting ringing; rather, the CNNs are used for reconstruction.
Similarly, Huang (Evaluation on the generalization of a learned convolutional neural network for MRI reconstruction. Magnetic Resonance Imaging. Volume 87. Pages 38-46. ISSN 0730-725X.), also teaches using CNNs for image reconstruction, and Huang investigates using separate CNNs for handling different sampling patterns [Fig. 4]. However, Huang does not teach using separate CNNs for correcting ringing in a one-dimensional direction and a two-dimensional direction for each sampling pattern. Rather, Huang concluded that using separate CNNs for handling different sampling patterns did not provide a significant advantage over using a single CNN trained on all sampling patterns [Section 3].
Thus, the prior art overall teaches using CNNs for MRI reconstruction and denoising, and the prior art motivates using separate CNNs for handling images of different sampling patterns as shown in the rejections to claims 1-3 and 7-8. However, the prior art lacks detailed implementations regarding using separate CNNs for ringing correction in a one-dimensional or a two-dimensional direction.
Regarding claim 5, the closest prior art of record, Zhang and Wuelker, teach the magnetic resonance imaging apparatus according to claim 1. However, both Zhang and Wuelker fail to teach wherein, in a case in which the measurement data is three-dimensional data obtained by radial scanning, the one or more processors select a combination of a CNN for a cylindrical or spherical sampling pattern or a CNN for an elliptical sampling pattern and a CNN for a rectangular sampling pattern.
Zhang teaches utilizing CNNs for performing ringing correction in MRI images, but Zhang teaches using different CNNs based on different under-sampling ratios and not based on different sampling patterns. Wuelker teaches utilizing CNNs for removing noise and artifacts (which can include ringing) from MRI images, and Wuelker teaches using different CNNs specifically trained to handle different sampling patterns. Wuelker mentions selecting a specific CNN for denoising images with a radial sampling pattern [0065]. However, Wuelker does not mention ringing correction of three-dimensional data by selecting a combination of different CNNs.
As discussed above regarding claim 4, the other prior art of record teaches using separate CNNs to handle different sampling patterns. However, the other prior art of record also fails to teach selecting a combination of these CNNs for handling 3D data.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Ruan et al. (Gibbs-Ringing Artifact reduction in MR images with varying sampling levels Via a Single Convolutional Neural Network. Proc. Intl. Soc. Mag. Reson. Med. 27.) teaches methods for using a single CNN to remove Gibbs ringing from MRI images. This single CNN is used for different sampling levels.
Huang et al. (Evaluation on the generalization of a learned convolutional neural network for MRI reconstruction. Magnetic Resonance Imaging. Volume 87. Pages 38-46. ISSN 0730-725X.) teaches methods of using DC-CNNs for MRI image reconstruction. Fig. 4 shows a performance comparison between different DC-CNNs which are trained to handle a specific sampling pattern type. In Section 3, Huang concludes that using separate DC-CNNs for different sampling patterns provides only a 0.02 increase in SSIM compared to using a single DC-CNN trained on all sampling patterns [Section 3].
Lebel (US 2020/0126190 A1) teaches systems and methods for using a deep CNN to identify different types of artifacts, including Gibbs ringing, in MRI images, so the artifacts can be removed. The CNN learns from MRI image data of different sampling patterns.
Fessler et al. (US 2023/0236271 A1) teaches systems and methods for using machine learning for MRI image reconstruction and denoising. The methods teach different implementations of image reconstructors, including a CNN denoiser for removing ringing, a model-based deep leaning image reconstructor, etc.
Li et al. (US 2019/0244399 A1) teaches systems and methods for removing Gibbs artifacts from MRI images, and these methods involve utilizing a CNN for receiving MRI data containing artifacts and producing an MRI image with artifacts removed.
Chen et al. (Deep Learning for Image Enhancement and Correction in Magnetic Resonance Imaging—State-of-the-Art and Challenges. J Digit Imaging 36, 204–230.) teaches methods for utilizing DNNs for artifact corrections. These methods utilize a DNN which receives an MRI image containing Gibbs ringing, and the DNN produces an artifact-free image.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ERIC JAMES SHOEMAKER whose telephone number is (571)272-6605. The examiner can normally be reached Monday through Friday from 8am to 5pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner' s supervisor, JENNIFER MEHMOOD, can be reached at (571)272-2976. The fax phone number for the organization where this application or proceeding is assigned is (571)273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Eric Shoemaker/
Patent Examiner
/XIAO LIU/Primary Examiner, Art Unit 2664
Prosecution Insights