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 .
Response to Amendment
Claims 1, 18-22, 25 has been amended.
Claims 16 has been canceled.
Claims 1, and 17-26 are still pending for consideration.
Response to Arguments
Applicant on pages 2-7 of the “Remarks” asserts “First, in the combination of Chan and Lebel, neither reference provides 1.) the missing prior, which is a statistical model and "realizations from a prior", 2.) the model images, 3.) the generative training sets, nor 4.) any motivation to replace real training data with synthetic, prior based data”.
Neither Chan nor Lebel Discloses Model Images Sampled from a Prior p(m)
As amended claim 1 recites "obtaining a model image mss* from a sample M*, wherein the model image mss* is a realization from a prior p(m), the prior p(m) is a statistical model based on expert prior data..." This limitation is also not taught or suggested in either Chan or Lebel, alone or in combination.
Response A:
Examiner respectfully disagree with applicant’s argument. Label on para [0048] disclose “the corrupted medical images may be synthesized images. For example, noise can be added to a clean image to generate the synthesized image. The specific type of noise depends on the imaging modality. The deep learning network is trained to map the synthesized image onto the additive noise. [0049] As another example, Gibbs ringing can be introduced to a high-resolution image by downsampling the image”. Label further on para [0037] disclose “The specific type of noise may depend on the imaging modality. Mathematically, this corresponds to: I.sub.s=I+ϵ, where I.sub.s is the synthesized image with additive noise, I is the clean image, and ε is the additive noise. In MRI, for example, images may be corrupt with complex-valued, white, Gaussian noise. In magnitude or ultrasound images, the noise may comprise Rician noise. In CT imaging, the noise may comprise Poisson noise. To that end, the additive noise ε may be modeled according to Gaussian, Rician, and Poisson distributions for MR, ultrasound, and CT imaging modalities respectively”). Thus, training a model using simulated or statistically generated degradation rather than heavily relying on purely real-world corrupted data, model-image mss corresponds to synthesized image, sample image M* corresponds to clean image, prior p(m) corresponds under broadest reasonable interpretation to imaging modality noise specific type of noise depends on imaging modality. Creating the synthetic data is heavily governed by expert domain knowledge of how MRI, CT, or X-Ray physics creates artifacts and model mapped to expert prior correspond to mapping onto additive noise; the deep learning network compares the synthesized, corrupted image to the clean image and learns to extract or separate the corruption. Therefore, applicant’s arguments are not persuasive.
Applicant additionally asserts “Neither Chan nor Lebel Discloses Selecting Expected Outputs Based on Model Images. As amended claim 1 recites, "...selecting the expected output mf* based on the model image mss* for each sim data image..." This limitation is also not taught or suggested by Chan or Level, alone or in combination.
Response B:
Examiner respectfully disagree with applicant’s argument. Label on para [0038] disclose “The synthesized image I.sub.s is used as input image 102, and the additive noise ε is used as the output 104 to train the deep learning network 200”. Using a synthesized image as the input and the corresponding additive noise (the difference between a clean image and a noisy/synthetic one) as the expected output. Therefore, synthesized Image is corresponding to sim-data-image, and additive noise corresponds expected output. Therefore, applicant’s arguments are not persuasive.
C. Applicant additionally asserts “Neither Chan nor Lebel Teaches Training a Machine Learning System Using Only Prior Derived Synthetic Data. As amended claim 1 recites, "applying the training sets as input for training the machine learning system... and training the machine learning system until the machine learning system converges based on comparing the outputs mo with the expected outputs mf*." This training occurs on synthetic model image derived sim data images, and expected outputs derived from the same model images. This limitation is also not taught or suggested by Chan or Level, alone or in combination.
Response C:
Examiner respectfully disagree with applicant’s argument. Label on para [0045] disclose “a loss function is used to iteratively adjust parameters (e.g., weights and biases of convolutional and pooling layers) of DL-CNN network until stopping criteria are satisfied (e.g., convergence of the parameters to a predefined threshold) to generate the trained network 162. The loss function compares high-quality data 153 to results of a current version of the DL-CNN network to which low-quality data 155 is applied”. Label further on para [0097] disclose “These predictions are then input into the loss function, by which they are compared to the corresponding ground truth labels (i.e., the high quality image 153)”. Label explicitly identifies the input as "low-quality data 155" and the output as "predictions, defines this mathematically as a Loss Function that pits the network predictions against the "ground truth" (the high-quality image 153), and how that convergence is achieved: by "iteratively adjusting parameters (weights and biases)" via backpropagation based on the loss function results. Therefore, applicant’s arguments are not persuasive.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1, 18 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Chan et al. (US 20190365341 A1) in view Lebel (US 20200126190 A1).
Regarding claim 1, Chan et al. teaches a computer-implemented method for image analysis and reconstruction for medical imaging of a subject, the method is comprising:- obtaining a PET image (see Abstract; “A deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images”, see also para [0019]; “FIG. 8A shows another example of a noisy PET image that has been reconstructed”) and dividing the PET image into a plurality of subset-images dsso, wherein each subset-image dsso is a representation of one, or more, pixels (see Fig. 4, Abstract; “slicing a three-dimensional (3D) PET image into 2D slices along transaxial, coronal, and sagittal planes” Note: 2D slices made up of pixels), providing a trained machine learning system (see para [0042]; “the trained DL-CNN can be used with input images spanning a wide range of statistical properties”), applying the subset-images dsso as input for the trained machine learning system (see Abstract; “using three separate 2D CNN networks for each respective plane, and averaging the outputs from these three separate 2D CNN networks”, see also para [0062]; “the transaxial network 422 can be applied to 2D slices that are parallel to a transaxial plane of the 3D PET image. Similarly, the coronal network 424 can be applied to 2D slices that are parallel to a coronal plane of the 3D PET image, and the sagittal network 426 can be applied to 2D slices that are parallel to a sagittal plane of the 3D PET image” Note: the sliced subset-images are fed to the trained networks), obtaining a output mo for each subset-image dsso from the trained machine learning system (see para [0033]; “The DL-CNN network described herein can produce output images having a uniform image quality, even when the input images exhibit large variations in their image quality and statistical properties (e.g., by having different noise levels)”, see also para [0064]; “the respective DL-CNN networks 422, 424, and 426 are applied to slices in their respective planes to generate the denoised images 430”), and determining and outputting a representation output based on the outputs mo (see para [0065]; “the denoised images are combined to form an aggregate image. For example, the three denoised images 430 can be averaged, resulting in a single 3D denoised image”, see also para [0118]; “The outputs of these collections can then tiled so that they overlap, to obtain a better representation of the original image”), and-training the machine learning system until the machine learning system converges based on comparing the outputs mo with the expected outputs mf* (see para [0045]; “a loss function is used to iteratively adjust parameters (e.g., weights and biases of convolutional and pooling layers) of DL-CNN network until stopping criteria are satisfied (e.g., convergence of the parameters to a predefined threshold) to generate the trained network 162. The loss function compares high-quality data 153 to results of a current version of the DL-CNN network to which low-quality data 155 is applied”, see also para [0097]; “These predictions are then input into the loss function, by which they are compared to the corresponding ground truth labels (i.e., the high quality image 153)”). However, Chan et al. does not teach wherein providing a trained machine learning system comprises:- obtaining a model-image mss* from a sample M*, wherein the model-image mss* is a realization from a prior p(m), the prior p(m) is a statistical model based on expert prior data, and each model-image mss* is a representation of an image of one, or more, pixels, -obtaining training sets, wherein each training set comprises a sim-data-image dss,sim* and an expected output mf*, by determining the sim-data-image dss,sim* based on the model-image mss*, and selecting the expected output mf* based on the model-image mss* for each sim-data-image dss,sim*, wherein each sim-data-image dss,sim* is a representation of one, or more, pixels, -applying the training sets as input for training the machine learning system, and obtaining a output mo for each sim-data-image dss,sim* from the machine learning system.
In the same field of endeavor, Label teaches wherein providing a trained machine learning system comprises:- obtaining a model-image mss* from a sample M*, wherein the model-image mss* is a realization from a prior p(m), the prior p(m) is a statistical model based on expert prior data (see para [0048]; “the corrupted medical images may be synthesized images. For example, noise can be added to a clean image to generate the synthesized image. The specific type of noise depends on the imaging modality. The deep learning network is trained to map the synthesized image onto the additive noise. [0049] As another example, Gibbs ringing can be introduced to a high resolution image by downsampling the image”, and para [0037]; “The specific type of noise may depend on the imaging modality. Mathematically, this corresponds to: I.sub.s=I+ϵ, [0038] where I.sub.s is the synthesized image with additive noise, I is the clean image, and ε is the additive noise. In MRI, for example, images may be corrupt with complex-valued, white, Gaussian noise. In magnitude or ultrasound images, the noise may comprise Rician noise. In CT imaging, the noise may comprise Poisson noise. To that end, the additive noise ε may be modeled according to Gaussian, Rician, and Poisson distributions for MR, ultrasound, and CT imaging modalities respectively”) Note; training a model using simulated or statistically generated degradation rather than heavily relying on purely real-world corrupted data, model-image mss corresponds to synthesized image, sample image M* corresponds to clean image, prior p(m) corresponds to imaging modality noise, and model mapped to expert prior correspond to mapping onto additive noise; the deep learning network compares the synthesized, corrupted image to the clean image and learns to extract or separate the corruption), and each model-image mss* is a representation of an image of one, or more, pixels (see para [0032]; “The input image 102 may be of any appropriate size, for example, 128×128 pixels, 256×256 pixels, 512×512 pixels, and so on”) -obtaining training sets, wherein each training set comprises a sim-data-image dss,sim* and an expected output mf*, by determining the sim-data-image dss,sim* based on the model-image mss* (see para [0035]; “The deep learning network 200 may be trained using corrupted medical images and artifacts present in corresponding images. In some embodiments, the corrupted mecical images may be synthesized images. For example, one or more types of artifacts may be added to a medical image with good quality (e.g., relatively clean image) to obtain a synthesized corrupted image. The synthesized corrupted image is used as input image 102, and the added one or more types of artifacts as outputs 104 and 106 to train the deep learning network 200”), and selecting the expected output mf* based on the model-image mss* for each sim-data-image dss,sim* (see para [0038]; “The synthesized image I.sub.s is used as input image 102, and the additive noise ε is used as the output 104 to train the deep learning network 200” Note; Synthesized Image is corresponding to sim-data-image, and additive noise corresponds expected output), wherein each sim-data-image dss,sim* is a representation of one, or more, pixels (see para [0050]; “The medical image may be of any appropriate size, for example, 128×128 pixels, 256×256 pixels, 512×512 pixels, and so on”), -applying the training sets as input for training the machine learning system, and obtaining a output mo for each sim-data-image dss,sim* from the machine learning system (see para [0035]; “The synthesized corrupted image is used as input image 102, and the added one or more types of artifacts as outputs 104 and 106 to train the deep learning network 200”, see also para [0038]; “The synthesized image I.sub.s is used as input image 102, and the additive noise ε is used as the output 104 to train the deep learning network 200”, and para [0046]; “one or more deep learning networks are trained to map corrupted images onto a first type and a second type of artifacts present in corresponding corrupted images”). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label in order to identify new ways for intelligently improving the acquired image quality (see para [0048]).
Regarding claim 18, the rejection of claim 1 is incorporated herein.
Label in the combination further teach wherein the sim-data-image dss,sim* is determined from the model-image mss* by the function of the type dss,sim* = g(mss*) + n(mss*) and, wherein g is a smoothing function and n is a noise function (see para [0035]; “the corrupted medical images may be synthesized images. For example, one or more types of artifacts may be added to a medical image with good quality (e.g., relatively clean image) to obtain a synthesized corrupted image”, see also para [0037]; “noise can be added to a clean image to generate the synthesized image. The specific type of noise may depend on the imaging modality. Mathematically, this corresponds to: I.sub.s=I+ϵ, [0038] where I.sub.s is the synthesized image with additive noise, I is the clean image, and ε is the additive noise”, and para [0036]; “the artifacts may include..additional blurring, etc.” Note: generate synthesized corrupted image from clean/model image (dss,sim* from mss*), noise function (n(m) and applying additional blurring (g(mss)) to the clean image implies smoothing function -i.e., dss,sim* = g(mss*) + n(mss*).
Regarding claim 26, the rejection of claim 1 is incorporated herein.
Chen et al. in the combination further teaches wherein the machine learning system comprises a neural network (see Abstract; “A deep learning (DL) convolution neural network (CNN) reduce noise in positron emission tomography (PET) images”).
Claims 17, and 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Chan et al. in view Label as applied in claim 1 above, and further in view of De Fauw et al. (US 20190005684 A1) herein after De.
Regarding claim 17, the rejection of claim 1 is incorporated herein.
The combination of Chan et al. and Label does not teach wherein providing a trained machine learning system comprises: - selecting the trained machine learning system from a plurality of trained machine learning systems based on the type of the output mo to be obtained.
In the same field of endeavor De teach wherein providing a trained machine learning system comprises: - selecting the trained machine learning system from a plurality of trained machine learning systems based on the type of the output mo to be obtained (see para [0004]; “implement a first set of one or more segmentation neural networks. Each segmentation neural network in the first set may be configured…to process the input image to generate a segmentation map … implement a set of one or more classification neural networks. wherein each classification neural network is configured to … process the classification input to generate a classification output”, see also para [0011]; “at least one first, image segmentation neural network, and at least one second, classification neural network” Note: the trained system uses the segmentation networks when segmentation map is the output and uses the classification networks when classification output is desired). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and method for generating a final classification output for an image of eye tissue of De in order to give the user an intuitive insight into segmentation confidence in different tissue regions, particularly in difficult or ambiguous cases (see para [0004]).
Regarding claim 23, the rejection of claim 1 is incorporated herein.
De in the combination further teach wherein the output mo is a numerical value representative of a pixel intensity or a mean value and a covariance for a number of pixels, or a number of pixels, or the expected output mo is a category or a probability for one or more categories (see para [0004]; “process the input image to generate a segmentation map that segments the eye tissue in the input image into a plurality of tissue types”, see also para [0044]; “(i) the voxel volume, and (ii) the number of voxels assigned to the particular tissue by the particular segmentation map”, and para [0071]; “the classification output … includes a respective condition score for each of multiple medical conditions. Each condition score may represent a predicted likelihood that the patient has the medical condition” Note: the system outputs segmentation (categorical label at the pixel/voxel level) and classification score/likelihood (probabilities)). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for generating a final classification output for an image of eye tissue of De in order to generate pathologies and/or clinical referral decisions (see para [0004]).
Regarding claim 24, the rejection of claim 1 is incorporated herein.
De in the combination further teach wherein the outputs mo are ordered into a representation output (see Fig. 2, see para [0008]; “The subsystem is further configured to provide a representation, for example a visualization, of at least one of the segmentation maps for presentation on a user device. Such an intermediate data output can provide an explanation of the final classification”, see also para [0056]; “After generating the respective classification outputs for each of the segmentation maps, the system 100 can combine the classification outputs to generate a final classification output 106. For example, the system 100 may determine the final classification output 106 as an average of the classification outputs for each of the segmentation maps”, and see para [0076]; “the system can provide the final classification output to a user device (212)”). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for generating a final classification output for an image of eye tissue of De in order to provide a tool which allows a clinician to make more informed diagnoses (see para [0008]).
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Chan et al. in view Lebel as applied in claim 1 above, and further in view of Lin et al. (US 8655109 B2).
Regarding claim 19, the rejection of claim 1 is incorporated herein. The combination of Chan et al. and Lebel does not teach wherein the selected expected output mf* is the pixel intensity of the central pixel, or a group of central pixels in the model-image mss*.
Lin et al. teach wherein the selected expected output mf* is the pixel intensity of the central pixel, or a group of central pixels in the model-image mss* (see col. 5, lines 31-34; “denoising the center pixel of the low resolution patch by averaging it with center pixels of the best matching patches, and performing regression for each of the best matching patches to generate a set of regression result”). (see col. 5, lines 31-34). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for a regression-based learning model in image upscaling of Lin et al. in order to avoid border effects and align the receptive field (see para [0008]).
Claims 20-21, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Chan et al. in view of Lebel as applied in claims 1, and 16, and further in view of De.
Regarding claim 20, the rejection of claim 16 is incorporated. The combination of Chan et al. and Lebel as a whole does not teach wherein the selected expected output mf* is the number of pixels connected to the central pixel in the model-image mss* with an intensity higher than a threshold intensity value or a probability of a disease is higher than a threshold probability value.
De teaches wherein the selected expected output mf* is the number of pixels connected to the central pixel in the model-image mss* with an intensity higher than a threshold intensity value or a probability of a disease is higher than a threshold probability value (see para[0006]; “according to a defined compromise between sensitivity of the system to making a final classification and a false alarm rate for the final classification, for example by rescaling classification probabilities from an ensemble of the classification neural networks”, see also para [0064]; “a set of scores or pseudo-probabilities, q, one for each class. These may be further processed to adjust a balance between accuracy (that is “sensitivity” or whether or not a classification is correct)…. The scaling factor a may be chosen such that a 50% pseudo-probability achieves a maximal (sensitivity+specificity)/2” Note: the system outputs per-condition probabilities (scores or pseudo-probabilities), setting a decision rule like p>threshold and mf* defined as a disease probability with a threshold). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for generating a final classification output for an image of eye tissue of De in order to encode local structure with reasonable expectation of success (see para [0006]).
Regarding claim 21, the rejection of claim 16 is incorporated.
De in the combination further teach wherein the expected output mf* is a category of the central pixel, or a group of central pixels, in the model-image mss* or the expected output mrf* is a probability for one or more categories (see para [0010]; “process the input image to generate a segmentation map that segments the eye tissue in the input image into a plurality of tissue types”, see also para [0040]; “each segmentation map assigns a respective tissue type from the predetermined set of tissue types to each voxel of the medical image 102”, and para [0048]; “Each condition score may represent a predicted likelihood that the patient 104 has the medical condition, conditioned on the segmentation map of the medical image 102 of the patient” Note: per-pixel/voxel categorical labels via segmentation maps (covering the central pixel or group)). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for generating a final classification output for an image of eye tissue of De in order to represent a state of medical condition and progress to the condition state at a particular future time (see para [0010]).
Regarding claim 25, the rejection of claim 16 is incorporated.
De in the combination further teach wherein the machine learning system comprises a regression type mapping, when the expected output mf* represents a numerical value, or a classification type mapping, when the expected output mf* represents a category (see para [0004]; “process the input image to generate a segmentation map that segments the eye tissue in the input image into a plurality of tissue types…a classification input derived from a segmentation map of eye tissue, and process the classification input to generate a classification output that characterizes the eye tissue”, see also para [0039]; “The segmentation module can generate the segmentation maps in any appropriate manner, for example,..linear regression engines, or a combination thereof”). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a method for generating a final classification output for an image of eye tissue of De in order to generate different hypotheses characterizing the medical image (see para [0004]).
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Chan et al. in view of Lebel as applied in claim 1 and further in view of Marks et al. (US 20210104068 A1).
Regarding claim 22, the rejection of claim 16 is incorporated. The combination of Chan et al. and Lebel as a whole does not teach wherein the expected output mf* is a vector with two values, the mean and the covariance for a normal distribution.
Marks et al. teach wherein the expected output mf* is a vector with two values, the mean and the covariance for a normal distribution (see para [0012]; “An example of a parametric probability distribution defined by values of parameters for a location of each landmark in each processed image is a Gaussian distribution, wherein the parameters determine a mean and a covariance matrix of the Gaussian distribution”, see also para [0019]; “the estimated landmark location is used as the mean of the Gaussian distribution, and the estimated covariance matrix is used as the covariance of the Gaussian distribution”). Accordingly, it would have been obvious to one of ordinary skill in the art before the invention of the claimed invention to modify a method of deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images of Chan et al. in view of the use of a deep learning network provided for selectively denoising medical images of Label and a neural network for executing a task based on probabilistic image-based landmark localization of Marks et al. in order to concurrently estimate the landmark locations and their uncertainties as well as improve the accuracy of the landmark localization (see para [0012]).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WINTA GEBRESLASSIE whose telephone number is (571)272-3475. The examiner can normally be reached Monday-Friday9:00-5:00.
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, Andrew Bee can be reached at 571-270-5180. 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.
/WINTA GEBRESLASSIE/Examiner, Art Unit 2677
Prosecution Insights