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 .
This is in response to the applicant’s reply filed April 7, 2026. In the applicant’s reply; claims 2, 12 and 17-18 were amended. Claims 14, 16 and 20 were cancelled. Claims 1-13, 15, 17-19 and 21-23 are pending in this application.
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 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.
Priority
This application repeats a substantial portion of prior Application No. 17/228690, filed February 8, 2024, and adds disclosure not presented in the prior application. Because this application names the inventor or at least one joint inventor named in the prior application, it may constitute a continuation-in-part of the prior application. Should applicant desire to claim the benefit of the filing date of the prior application, attention is directed to 35 U.S.C. 120, 37 CFR 1.78, and MPEP § 211 et seq. The presentation of a benefit claim may result in an additional fee under 37 CFR 1.17(w)(1) or (2) being required, if the earliest filing date for which benefit is claimed under 35 U.S.C. 120, 121, 365(c), or 386(c) and 1.78(d) in the application is more than six years before the actual filing date of the application.
Examiner’s Responses to Applicant’s Remark
Applicants' amendments filed on April 7, 2026 have been fully considered. The amendments overcome the following rejections set forth in the office action mailed on January 15, 2026.
Applicant’s amendments overcome the objection to the title of the specification, and the objection is hereby withdrawn.
Applicants' arguments filed on April 7, 2026 have been fully considered but they are not persuasive. The Examiner has thoroughly reviewed Applicants' arguments but firmly believes that the cited reference to reasonably and properly meet the claimed limitation.
Applicant argues “that Claim 1 of US Patent 11,90046 recites a training process of a model…while claim 1 of the present application is directed to the application of a trained model”.
Examiner respectfully disagrees. Both sets of claims have features directed to the training and usage of the application, and both sets of claims input in a first projection data, generate a second projection data, and use that to train and output a correction coefficient as a processing parameter as exemplified by independent claim 1, presented below, as exemplary, of the overall embodiments. Both applications appear to be co-extensive in scope. As a result, Claims 1-20 are still rejected on the ground of nonstatutory double patenting as being unpatentable over claims 1-20 of U.S. Patent No. 11,908,046 issued from US Application 17/228,690, and the rejection is maintained and presented below.
Instant Application 18/437,210
US Patent 11,908,046 issued from 17/228,690
1. A method implemented on at least one computing device, each of which has at least one processor and storage device, the method comprising:
obtaining first projection data of a first subject, wherein the first projection data is acquired by a first medical device;
generating second projection data of the first subject based on the first projection data of the first subject;
inputting the first projection data and the second projection into a trained model; and outputting a processing parameter by the trained model processing the first projection data and second projection data, the processing parameter including a correction coefficient for correcting errors introduced by the first medical device.
1. A method implemented on at least one computing device, each of which has at least one processor and storage device, the method comprising:
obtaining first projection data of a subject;
generating second projection data of the subject corresponding to the first projection data of the subject;
and obtaining a trained model, by training an initial model with both the first projection data and the second projection data as an input of the initial model, wherein an output of the trained model is a processing parameter, and the processing parameter includes a correction coefficient for correcting errors introduced by a detector.
The nature of the claimed features in both applications and the manner in which they are recited do make them co-extensive in scope and are there subject to the double patenting rejections of record, presented below.
Applicant argues that Suzuki teaches the output of a high quality image and does not necessarily teach the output of a “processing parameter” or a “correction coefficient” as claimed in the independent claims.
Examiner respectfully disagrees. Applicants are reminded that the Examiner is entitled to give the broadest reasonable interpretation to the language of the claims. The claims in the instant application require the following feature in independent claims 1 and 12: in [0021] Suzuki teaches “When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used.” So the Examiner considers the output of a linear BP algorithm to be Applicants' “processing parameter” or “correction coefficient” within the broad meaning of the term. The Examiner is not limited to Applicants' definition which is not specifically set forth in the claims. In re Tanaka et al., 193 USPQ 139, (CCPA) 1977.
Applicant argues that “the input of Feng is a small amount of projection data” and the output of the model is a CT image”
Applicants are reminded that the Examiner is entitled to give the broadest reasonable interpretation to the language of the claims. The claimed limitations require the input of projection data, and as Feng’s teachings are directed to this type of input, as acknowledged by applicant in the arguments presented. Likewise, Feng has an entire section dedicated to “Numerical Experiments of the CT image reconstruction, subsection 3.1. Model problems, which teaches a process for evaluating a reconstruction error: “In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers” and further teaches the overall weighting process. So the Examiner considers the error values determined by Feng to be Applicants' the processing parameter including a correction coefficient for correcting errors within the broad meaning of the term. The Examiner is not limited to Applicants' definition which is not specifically set forth in the claims. In re Tanaka et al., 193 USPQ 139, (CCPA) 1977.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claims 1-20 are rejected on the ground of nonstatutory double patenting as being unpatentable over claims 1-20 of U.S. Patent No. 11,908,046 issued from US Application 17/228,690. Although the claims at issue are not identical, they are not patentably distinct from each other because they are both directed towards training a model for generating a processing parameter for first and second projection data to determine a correction coefficient.
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 may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Suzuki et al. (US PGPub US 2015/0201895 A1, hereby referred to as “Suzuki”), in view of Feng et al. (Neural network CT image reconstruction method for small amount of projection data, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, Volume 449, Issues 1–2, 2000,Pages 366-377,), hereby referred to as “Feng”.
Consider Claims 1, 12 and 17.
Suzuki teaches:
1. A method implemented on at least one computing device, each of which has at least one processor and storage device, the method comprising: / 12. A system for determining a parameter for medical data processing, comprising: / 17. A method implemented on at least one computing device, each of which has at least one processor and storage for determining a parameter for medical data processing, the method comprising: (Suzuki: abstract, [0014]-[0021], Figures 1-3, [0014] FIG. 1 shows both a training phase 102, in which the supervised dose reduction technique is developed for image conversion, and a post-training, application phase 104, in which the learned image conversion is used. In all, the techniques allow for the use of lower, more acceptable radiation dosages for CT imaging. [0017] FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining “input” medical images (stage 202) and “teaching” medical images (stage 204), respectively.)
12. at least one storage medium including a set of instructions; and at least one processor configured to communicate with the at least one non-transitory storage medium, wherein when executing the set of instructions, the at least one processor is directed to: (Suzuki: [0019]-[0020], [0031]-[0033], Figure 7, [0031] With reference to FIG. 7, an exemplary system for implementing the blocks of the claimed method and apparatus includes a general-purpose computing device in the form of a computer 12. Components of computer 12 may include, but are not limited to, a processing unit 14 and a system memory 16.)
1. obtaining first projection data of a first subject, wherein the first projection data is acquired by a first medical device; / 12. obtain first projection data of a first subject, wherein the first projection data is acquired by a first medical device; / 17. obtaining projection data and at least one scanning parameter, wherein the projection data is generated by a scanner under the at least one scanning parameter; (Suzuki: [0017] FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining "input" medical images (stage 202) and "teaching" medical images (stage 204), respectively. Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts.)
1. generating second projection data of the first subject based on the first projection data of the first subject; / 12. generate second projection data of the first subject based on the first projection data of the first subject; (Suzuki: [0017] FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining "input" medical images (stage 202) and "teaching" medical images (stage 204), respectively. Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts.)
1. inputting the first projection data and the second projection into a trained model; / 12. input the first projection data and the second projection into a trained model; / 17. obtaining a trained neural network model; (Suzuki: [0017] FIG. 2 Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts. [0021], Various pixel/voxel-based machine learning (PML) techniques may be applied as described herein, these include neural filters, neural edge enhancers, neural networks, shift-invariant neural networks, artificial neural networks (ANN), including massive-training ANN (MTANN), massive-training Gaussian process regression, and massive-training support vector regression (MTSVR), by way of examples. Additional techniques for error analysis and medical image data comparisons between an "input" image and a "training" image include those provided in U.S. Pat. Nos. 6,754,380, 6,819,790, and 7,545,965, and U.S. Publication No. 2006/0018524, the entire specifications of all of which are hereby incorporated by reference, in their respective entireties.)
1. and outputting a processing parameter by the trained model processing the first projection data and second projection data. / 12. and output a processing parameter by the trained model processing the first projection data and second projection data. / 17. inputting the projection data and the at least one scanning parameter into the trained neural network model: and outputting the parameter by the trained neural network model by processing the projection data and the at least one scanning parameter (Suzuki: [0021] A supervised dose reduction converter is trained, at a block 216, by using a training algorithm for the machine learning model developed at block 214. A training module 217, which may be stored in a non-transitory computer readable medium, such as a computer memory, for execution by a processor, as shown in FIG. 7, may perform the training of block 216. When the machine-learning model is a multi-layer perceptron, an error back-propagation (BP) algorithm can be used. When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used. After training, the supervised dose reduction converter (block 216) is able to assess sub-regions/sub-volumes of incoming non-training input images (from block 218) and convert those to output pixel/voxel values and resulting images (at block 220) similar to or close to the corresponding values as would appear in an HDCT image of the same corresponding structures. Thus, the supervised dose reduction technique acquires the function of converting LDCT images with noise and artifacts into HD like CT images with less noise or fewer artifacts, as in the illustrated examples.)
Even if Suzuki does not teach: 1/12. the processing parameter including a correction coefficient for correcting errors introduced by the first medical device / 17. the parameter comprising a correction coefficient
Feng teaches:
1. A method implemented on at least one computing device, each of which has at least one processor and storage device, the method comprising: / 12. A system for determining a parameter for medical data processing, comprising: at least one storage medium including a set of instructions; and at least one processor configured to communicate with the at least one storage medium, wherein when executing the set of instructions, the at least one processor is directed to: / 17. A method implemented on at least one computing device, each of which has at least one processor and storage for determining a parameter for medical data processing, the method comprising: (Feng: page 366, abstract, This paper presents a new method for two-dimensional image reconstruction by using a multi-layer neural network. Though a conventionally used object function of such a neural network is composed of a sum of squared errors of the output data, we define an object function composed of a sum of squared residuals of an integral equation. By employing an appropriate numerical line integral for this integral equation, we can construct a neural network which can be used for CT image reconstruction for cases with small amount of projection data. We applied this method to some model problems and obtained satisfactory results. This method is especially useful for analyses of laboratory experiments or held observations where only a small amount of projection data is available in comparison with the well-developed medical applications)
1. obtaining first projection data of a first subject, wherein the first projection data is acquired by a first medical device; / 12. obtain first projection data of a first subject, wherein the first projection data is acquired by a first medical device; / 17. obtaining projection data and at least one scanning parameter, wherein the projection data is generated by a scanner under the at least one scanning parameter; (Feng: page 367 section 2. Neural network CT image reconstruction algorithm 2.1. Projection data We represent a model object (a model distribution) to be reconstructed in the two-dimensional domain X as z=f (x, y). The projection data for the CT image reconstruction are the line integrals of z along projection paths (Fig. 1). There are P projection paths in total and the pth projection path is specified by two parameters rp and hp, where rp is a distance from the origin to the pth projection path, and hp is the angle of the perpendicular of the pth projection path from the x-axis. The line integral g(rp, hp) along the pth path is given as
PNG
media_image1.png
66
524
media_image1.png
Greyscale
)
1. generating second projection data of the first subject based on the first projection data of the first subject; / 12. generate second projection data of the first subject based on the first projection data of the first subject; (Feng: page 367 section 2.1 Projection Data As in practical situations projection data include noise generally, we add a noise term n(rp, hp) to the above ideal model projection data g(rp, hp) and obtain the
PNG
media_image2.png
204
504
media_image2.png
Greyscale
)
1. inputting the first projection data and the second projection into a trained model; / 12. input the first projection data and the second projection into a trained model; / 17. obtaining a trained neural network model; (Feng: page 368 In our method we make full use of the following features of the neural network: 1. As the mapping function from the input data to the output data is obtained as a result of the learning process, it is not necessary to prepare a function or a group of functions which "t the mapping function in advance. The learning is carried out as an optimization problem of an appropriately defined object function. 2. One of the most important features of the mapping function of the neural network is that it contains smoothing and interpolating functions in addition to the curve-fitting function [15]. Moreover, the mapping function is continuous and differentiable. 3. As a curve-fitting method the expressive power of the neural network seems considerably superior to that of the standard methods by the linear combinations of the orthogonal functions such as the Fourier expansion because in the case of the neural network the `basis functions as themselves are adjusted during the learning process for better curve-fitting [19].)
1. and outputting a processing parameter by the trained model processing the first projection data and second projection data, the processing parameter including a correction coefficient for correcting errors introduced by the first medical device. / 12. and output a processing parameter by the trained model processing the first projection data and second projection data, the processing parameter including a correction coefficient for correcting errors introduced by the first medical device. / 17. inputting the projection data and the at least one scanning parameter into the trained neural network model: and outputting the parameter by the trained neural network model by processing the projection data and the at least one scanning parameter (Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers. Page 368 section 2.2. Multi-layer neural network We employ a feedforward multi-layer neural network composed of many neurons aligned in layers where a single neuron in a layer is connected to all the neurons in the neighboring layers and data are transmitted from the input layer to the output layer via hidden layers. In each neuron weighted sum of input data are calculated and the result is transformed by a nonlinear activation function. As an example a three-layered neural network is depicted in Fig. 2 by which a set of input data xi (i"1,2, I) is mapped to a set of output data yk (k"1,2,K).)
It would have been obvious before the effective filing date of the claimed invention to one of ordinary skill in the art to modify Suzuki with the teachings of Feng as they are both directed towards the use of neural networks in radiation image analysis and reconstruction. The determination of obviousness is predicated upon the following findings: One skilled in the art would have been motivated to modify Suzuki’s method and system for neural network based CT reconstruction in order to incorporate Feng’s teachings of a neural network CT image reconstruction method using iterative reconstruction, smoothing functions and correction coefficients in an iterative manner to better handle small amount of projection data. Furthermore, the prior art collectively includes each element claimed (though not all in the same reference), and one of ordinary skill in the art could have combined the elements in the manner explained above using known engineering design, interface and programming techniques, without changing a “fundamental” operating principle of Suzuki, while the teaching of Feng continues to perform the same function as originally taught prior to being combined, in order to produce the repeatable and predictable result of improving the overall radiation based CT reconstruction to leverage a multi-layer neural network for dose reduction and computational efficiency. It is for at least the aforementioned reasons that the examiner has reached a conclusion of obviousness with respect to the claim in question.
Consider Claims 2 and 11.
The combination of Suzuki and Feng teaches:
2. The method of claim 1, wherein the correction coefficient is configured to correct third projection data of a second subject acquired by the first medical device, and the second subject is different from the first subject. / 11. The method of claim 1, wherein the correction coefficient is configured to correct errors in projection data collected by a detector of the first medical device. (Suzuki: [0017] FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining “input” medical images (stage 202) and “teaching” medical images (stage 204), respectively. Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts. The number of “input” images may be comparatively small, 1, 10 or less, or 100 or less, by way of example. The number of “training” images have be small as well, 1, 10 or less, 20, or 50 or less. However, a larger number of “training” images may be used as well, 100-1,000 images, 1,000-10,000 images, or more than 10,000 images. The number of training images used may be adjusted from a small number to a high number based on the size of the “input” image, the desired reduction in SNR on the converted “input” image, the desired resolution of the edge effects on the converted “input” image, the number of and variation in the likely edges in the “input” image, the desired signal contrast on the converted “input” image, the radiation dose of the “input” image, the number of prior CT scans of a patient (accumulated radiation dose level), and the processing load on the computer system performing the comparisons. [0022] FIG. 3 a illustrates an example of conversion of a non-training, input lower quality image, in this case an LDCT image. The image taken with a dosage of 0.1 mSv is characterized by relatively high noise, e.g., having a signal-to-noise ratio (SNR) of 4.2 dB and various spurious artifacts. [0021] A supervised dose reduction converter is trained, at a block 216, by using a training algorithm for the machine learning model developed at block 214. A training module 217, which may be stored in a non-transitory computer readable medium, such as a computer memory, for execution by a processor, as shown in FIG. 7, may perform the training of block 216. When the machine-learning model is a multi-layer perceptron, an error back-propagation (BP) algorithm can be used. When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used. After training, the supervised dose reduction converter (block 216) is able to assess sub-regions/sub-volumes of incoming non-training input images (from block 218) and convert those to output pixel/voxel values and resulting images (at block 220) similar to or close to the corresponding values as would appear in an HDCT image of the same corresponding structures. Thus, the supervised dose reduction technique acquires the function of converting LDCT images with noise and artifacts into HD like CT images with less noise or fewer artifacts, as in the illustrated examples. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers.)
Consider Claims 3 and 13.
The combination of Suzuki and Feng teaches:
3. The method of claim 2, wherein correcting the third projection data comprises: constructing a correction model based on the correction coefficient; and generating the corrected third projection data based on the third projection data and the correction model./ 13. The system of claim 12, wherein the correction coefficient is configured to correct third projection data of a second subject acquired by the first medical device. (Suzuki: [0021] A supervised dose reduction converter is trained, at a block 216, by using a training algorithm for the machine learning model developed at block 214. A training module 217, which may be stored in a non-transitory computer readable medium, such as a computer memory, for execution by a processor, as shown in FIG. 7, may perform the training of block 216. When the machine-learning model is a multi-layer perceptron, an error back-propagation (BP) algorithm can be used. When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used. After training, the supervised dose reduction converter (block 216) is able to assess sub-regions/sub-volumes of incoming non-training input images (from block 218) and convert those to output pixel/voxel values and resulting images (at block 220) similar to or close to the corresponding values as would appear in an HDCT image of the same corresponding structures. Thus, the supervised dose reduction technique acquires the function of converting LDCT images with noise and artifacts into HD like CT images with less noise or fewer artifacts, as in the illustrated examples. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers.)
Consider Claims 4
The combination of Suzuki and Feng teaches:
4. The method of claim 2, wherein the correction coefficient is configured to correct an artifact relating to the third projection data.(Suzuki: [0017] FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining “input” medical images (stage 202) and “teaching” medical images (stage 204), respectively. Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts. The number of “input” images may be comparatively small, 1, 10 or less, or 100 or less, by way of example. The number of “training” images have be small as well, 1, 10 or less, 20, or 50 or less. However, a larger number of “training” images may be used as well, 100-1,000 images, 1,000-10,000 images, or more than 10,000 images. The number of training images used may be adjusted from a small number to a high number based on the size of the “input” image, the desired reduction in SNR on the converted “input” image, the desired resolution of the edge effects on the converted “input” image, the number of and variation in the likely edges in the “input” image, the desired signal contrast on the converted “input” image, the radiation dose of the “input” image, the number of prior CT scans of a patient (accumulated radiation dose level), and the processing load on the computer system performing the comparisons. [0022] FIG. 3 a illustrates an example of conversion of a non-training, input lower quality image, in this case an LDCT image. The image taken with a dosage of 0.1 mSv is characterized by relatively high noise, e.g., having a signal-to-noise ratio (SNR) of 4.2 dB and various spurious artifacts.)
Consider Claims 5, 15 and 18-19.
The combination of Suzuki and Feng teaches:
5. The method of claim 1, wherein the first subject comprises a phantom. / 15. The system of claim 12, wherein the first subject comprises a phantom. / 18. The method of claim 17, wherein the at least one scanning parameters comprise at least one of a tube voltage of the scanner or a tube current of the scanner. / 19. The method of claim 17, wherein the projection data is generated by scanning air under the at least one scanning parameter. (Examiner Note: In CT imaging, phantoms which consist of air provide a reference value for imaging; Suzuki: [0027] To train the supervised dose reduction technique, i.e., final image converter in FIG. 2, 6 sets of CT images of a chest phantom (Kyoto Kagaku, Kyoto, Japan) were acquired with a tube voltage of 120 kVp, tube current of 10, 25, 50, 100, 150, and 300 mA, and a collimation of 5 mm. CT images were reconstructed with the lung reconstruction kernel. Each reconstructed CT image had a matrix size of 512×512 pixels with no overlap between slices. A 10 mA (0.1 mSv) ultra-ultra-LDCT image and the corresponding 300 mA (3 mSv) HDCT image were used for training the supervised dose reduction technique as the input image and teaching image, respectively. We evaluated the image quality of CT images using signal-to-noise ratio (SNR) in each image with use of corresponding 3 mSv HDCT images as the reference standard. [0028] To evaluate the generalizability of the supervised dose reduction technique, we acquired ultra-ultra-LDCT (ULDCT) scans of 3 human patients with a tube voltage of 120 kVp and a tube current of 10 mA. The effective radiation dose of an ULDCT study was 0.1 mSv. We evaluated the image quality of CT images by using signal-to-noise ratio (SNR) in each image. We applied the supervised dose reduction technique trained with the phantom to the patient cases. With the trained supervised dose reduction technique, noise and artifacts (e.g., streaks) in ULDCT images (0.1 mSv) were reduced substantially, while details of soft tissue such as pulmonary vessels and bones were maintained, as illustrated in FIGS. 4 a/4 b, 5 a/5 b, and 6 a/6 b. In these example implementation, the average SNR for the 0.1 mSv ULDCT images for patients was improved from 2.3 (±1.8) to 13.0 (±2.5) dB (two-tailed t-test; P<.05). This 10.7 dB average SNR improvement was comparable to the 11.5 dB improvement that we were able to achieve by increasing the effective radiation dose from 0.1 mSv (10 mA) to 1.5 mSv (150 mA) in the phantom study, used as a reference, as illustrated by comparing FIGS. 5 a/5 b and 6 a/6 b.)
Consider Claims 6 and 16.
The combination of Suzuki and Feng teaches:
6. The method of claim 1, wherein the trained model is generated by a process comprising: generating an initial neural network model; obtaining first sample projection data of a third subject, wherein the first sample projection data of the third subject is generated by scanning the third subject with a second medical device; generating second sample projection data of the third subject based on the first sample projection data of the third subject; and training the initial neural network model with the first sample projection data and the second sample projection data to obtain the trained model.
(Suzuki: [0026] In an example test implementation of the present techniques, instead of using real LDCT images, simulated LDCT images were used. For example, simulated LDCT images were formed by degrading real HDCT images, and using these degraded images as input images to the supervised dose reduction technique. The major noise in LDCT images was quantum noise. Simulated quantum noise (which can be modeled as signal-dependent noise) is added to high-radiation-dose sinograms, fO(ξ, φ), acquired at a high radiation dose level, represented by
PNG
media_image3.png
156
406
media_image3.png
Greyscale
amount of noise. Simulated low-radiation-dose sinograms obtained with this method used for creating simulated LDCT images by using a reconstruction algorithm such as filtered back projection or an iterative reconstruction algorithm. Similarly, HDCT images are reconstructed from original HD sinograms. Instead of the above quantum noise model alone, a more realistic stochastic noise model can be used. In addition to the quantum noise, the stochastic noise model may include energy-integrating detectors, tube-current modulation, bowtie beam filtering, and electronic system noise. Alternatively, simulated LDCT images can be obtained by using a LDCT simulator in a CT system. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers.))
Consider Claims 7.
The combination of Suzuki and Feng teaches:
7. The method of claim 6, wherein the generating the second projection data of the first subject based on the first projection data of the first subject comprises: generating the second projection data of the first subject by correcting the first projection data of the first subject. (Suzuki: [0021] A supervised dose reduction converter is trained, at a block 216, by using a training algorithm for the machine learning model developed at block 214. A training module 217, which may be stored in a non-transitory computer readable medium, such as a computer memory, for execution by a processor, as shown in FIG. 7, may perform the training of block 216. When the machine-learning model is a multi-layer perceptron, an error back-propagation (BP) algorithm can be used. When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used. After training, the supervised dose reduction converter (block 216) is able to assess sub-regions/sub-volumes of incoming non-training input images (from block 218) and convert those to output pixel/voxel values and resulting images (at block 220) similar to or close to the corresponding values as would appear in an HDCT image of the same corresponding structures. Thus, the supervised dose reduction technique acquires the function of converting LDCT images with noise and artifacts into HD like CT images with less noise or fewer artifacts, as in the illustrated examples. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers.)
Consider Claim 8.
The combination of Suzuki and Feng teaches:
8. The method of claim 1, wherein the generating second projection data of the first subject based on the first projection data of the first subject comprises: reconstructing a first image of the first subject from the first projection data of the first subject; smoothing the first image of the first subject to generate a second image of the first subject; and projecting the second image of the first subject to generate the second projection data of the first subject. (Suzuki: [0026] In an example test implementation of the present techniques, instead of using real LDCT images, simulated LDCT images were used. For example, simulated LDCT images were formed by degrading real HDCT images, and using these degraded images as input images to the supervised dose reduction technique. The major noise in LDCT images was quantum noise. Simulated quantum noise (which can be modeled as signal-dependent noise) is added to high-radiation-dose sinograms, fO(ξ, φ), acquired at a high radiation dose level, represented by
PNG
media_image3.png
156
406
media_image3.png
Greyscale
amount of noise. Simulated low-radiation-dose sinograms obtained with this method used for creating simulated LDCT images by using a reconstruction algorithm such as filtered back projection or an iterative reconstruction algorithm. Similarly, HDCT images are reconstructed from original HD sinograms. Instead of the above quantum noise model alone, a more realistic stochastic noise model can be used. In addition to the quantum noise, the stochastic noise model may include energy-integrating detectors, tube-current modulation, bowtie beam filtering, and electronic system noise. Alternatively, simulated LDCT images can be obtained by using a LDCT simulator in a CT system. Feng: page 368 In our method we make full use of the following features of the neural network: 1. As the mapping function from the input data to the output data is obtained as a result of the learning process, it is not necessary to prepare a function or a group of functions which "t the mapping function in advance. The learning is carried out as an optimization problem of an appropriately defined object function. 2. One of the most important features of the mapping function of the neural network is that it contains smoothing and interpolating functions in addition to the curve-fitting function [15]. Moreover, the mapping function is continuous and differentiable. 3. As a curve-"tting method the expressive power of the neural network seems considerably superior to that of the standard methods by the linear combinations of the orthogonal functions such as the Fourier expansion because in the case of the neural network the `basis functions as themselves are adjusted during the learning process for better curve-fitting [19].)
Consider Claim 9.
The combination of Suzuki and Feng teaches:
9. The method of claim 1, wherein the generating the second projection data of the first subject based on the first projection data of the first subject comprises: smoothing the first projection data of the first subject to generate the second projection data. (Feng: page 368 In our method we make full use of the following features of the neural network: 1. As the mapping function from the input data to the output data is obtained as a result of the learning process, it is not necessary to prepare a function or a group of functions which "t the mapping function in advance. The learning is carried out as an optimization problem of an appropriately defined object function. 2. One of the most important features of the mapping function of the neural network is that it contains smoothing and interpolating functions in addition to the curve-fitting function [15]. Moreover, the mapping function is continuous and differentiable. 3. As a curve-"tting method the expressive power of the neural network seems considerably superior to that of the standard methods by the linear combinations of the orthogonal functions such as the Fourier expansion because in the case of the neural network the `basis functions as themselves are adjusted during the learning process for better curve-fitting [19].)
Consider Claim 10.
The combination of Suzuki and Feng teaches:
10. The method of claim 1, wherein the generating the second projection data of the first subject based on the first projection data of the first subject comprises: reconstructing a first image of the first subject from the first projection data of the first subject; modelling the first subject according to the first image; and calculating analytic equations of an X-ray transmission process to obtain the second projection data of the first subject. (Suzuki: [0017] FIG. 2 FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining "input" medical images (stage 202) and "teaching" medical images (stage 204), respectively. Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts. [0021], Various pixel/voxel-based machine learning (PML) techniques may be applied as described herein, these include neural filters, neural edge enhancers, neural networks, shift-invariant neural networks, artificial neural networks (ANN), including massive-training ANN (MTANN), massive-training Gaussian process regression, and massive-training support vector regression (MTSVR), by way of examples. Additional techniques for error analysis and medical image data comparisons between an "input" image and a "training" image include those provided in U.S. Pat. Nos. 6,754,380, 6,819,790, and 7,545,965, and U.S. Publication No. 2006/0018524, the entire specifications of all of which are hereby incorporated by reference, in their respective entireties. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers. Page 368 section 2.2. Multi-layer neural network We employ a feedforward multi-layer neural network composed of many neurons aligned in layers where a single neuron in a layer is connected to all the neurons in the neighboring layers and data are transmitted from the input layer to the output layer via hidden layers. In each neuron weighted sum of input data are calculated and the result is transformed by a nonlinear activation function. As an example a three-layered neural network is depicted in Fig. 2 by which a set of input data xi (i"1,2, I) is mapped to a set of output data yk (k"1,2,K).)
Consider Claim 20. CANCELLED
Consider Claim 21
The combination of Suzuki and Feng teaches:
21. (New) The method of claim 1, wherein the second projection data is corrected projection data of the first projection data. (Suzuki: [0017] FIG. 2 FIG. 2 illustrates an example flow diagram of a process 200 for performing a supervised dose reduction technique, and showing two initial stages obtaining "input" medical images (stage 202) and "teaching" medical images (stage 204), respectively.Once the image types are obtained they may be provided to a supervised machine learning technique, as shown in FIG. 1, for converting lower quality images, e.g., LDCT images with noise and artifacts, into high quality images, e.g., HD-like CT images with less noise or fewer artifacts. [0021], Various pixel/voxel-based machine learning (PML) techniques may be applied as described herein, these include neural filters, neural edge enhancers, neural networks, shift-invariant neural networks, artificial neural networks (ANN), including massive-training ANN (MTANN), massive-training Gaussian process regression, and massive-training support vector regression (MTSVR), by way of examples. Additional techniques for error analysis and medical image data comparisons between an "input" image and a "training" image include those provided in U.S. Pat. Nos. 6,754,380, 6,819,790, and 7,545,965, and U.S. Publication No. 2006/0018524, the entire specifications of all of which are hereby incorporated by reference, in their respective entireties. Feng: page 367 section 2.1 Projection Data In our method we make use of the following features of the neural network…4. As the learning process is a kind of nonlinear optimization processes, it is necessary to repeat the learning iterations extremely many times in order to attain high accuracy as long as a first-order optimization algorithm is used. pages 370-372, section 3 Numerical Experiments of the CT image reconstruction, section 3.1 In order to evaluate the reconstruction error, we define the projection error E5 and the relative average function error; section 3.2. Neural network structure In order to determine the neural network structure, an attained error after a fixed iteration number is investigated for a test problem by varying the number of hidden layers. Page 368 section 2.2. Multi-layer neural network We employ a feedforward multi-layer neural network composed of many neurons aligned in layers where a single neuron in a layer is connected to all the neurons in the neighboring layers and data are transmitted from the input layer to the output layer via hidden layers. In each neuron weighted sum of input data are calculated and the result is transformed by a nonlinear activation function. As an example a three-layered neural network is depicted in Fig. 2 by which a set of input data xi (i"1,2, I) is mapped to a set of output data yk (k"1,2,K).)
Consider Claim 22.
The combination of Suzuki and Feng teaches:
22. (New) The method of claim 1, wherein an input of the trained model further includes a scanning parameter. (Suzuki: [0021] A supervised dose reduction converter is trained, at a block 216, by using a training algorithm for the machine learning model developed at block 214. A training module 217, which may be stored in a non-transitory computer readable medium, such as a computer memory, for execution by a processor, as shown in FIG. 7, may perform the training of block 216. When the machine-learning model is a multi-layer perceptron, an error back-propagation (BP) algorithm can be used. When the machine-learning module is a linear-output artificial neural network (ANN) regression (see, for example, Suzuki, Pixel-Based Machine Learning in Medical Imaging, International Journal of Biomedical Imaging, Vol. 2012, Article ID 792079 incorporated by reference herein), a linear output BP algorithm can be used. After training, the supervised dose reduction converter (block 216) is able to assess sub-regions/sub-volumes of incoming non-training input images (from block 218) and convert those to output pixel/voxel values and resulting images (at block 220) similar to or close to the corresponding values as would appear in an HDCT image of the same corresponding structures. Thus, the supervised dose reduction technique acquires the function of converting LDCT images with noise and artifacts into HD like CT images with less noise or fewer artifacts, as in the illustrated examples.) (Feng: page 367 section 2. Neural network CT image reconstruction algorithm 2.1. Projection data We represent a model object (a model distribution) to be reconstructed in the two-dimensional domain X as z=f (x, y). The projection data for the CT image reconstruction are the line integrals of z along projection paths (Fig. 1). There are P projection paths in total and the pth projection path is specified by two parameters rp and hp, where rp is a distance from the origin to the pth projection path, and hp is the angle of the perpendicular of the pth projection path from the x-axis. The line integral g(rp, hp) along the pth path is given as
PNG
media_image1.png
66
524
media_image1.png
Greyscale
)
Consider Claim 23.
The combination of Suzuki and Feng teaches:
23. (New) The method of claim 19, wherein the parameter is used for correcting projection data acquired by scanning an object under the at least one scanning parameter, and the object includes at least one of a human body, a part of human body, an animal, or a phantom. (Examiner Note: In CT imaging, phantoms which consist of air provide a reference value for imaging; Suzuki: [0027] To train the supervised dose reduction technique, i.e., final image converter in FIG. 2, 6 sets of CT images of a chest phantom (Kyoto Kagaku, Kyoto, Japan) were acquired with a tube voltage of 120 kVp, tube current of 10, 25, 50, 100, 150, and 300 mA, and a collimation of 5 mm. CT images were reconstructed with the lung reconstruction kernel. Each reconstructed CT image had a matrix size of 512×512 pixels with no overlap between slices. A 10 mA (0.1 mSv) ultra-ultra-LDCT image and the corresponding 300 mA (3 mSv) HDCT image were used for training the supervised dose reduction technique as the input image and teaching image, respectively. We evaluated the image quality of CT images using signal-to-noise ratio (SNR) in each image with use of corresponding 3 mSv HDCT images as the reference standard. [0028] To evaluate the generalizability of the supervised dose reduction technique, we acquired ultra-ultra-LDCT (ULDCT) scans of 3 human patients with a tube voltage of 120 kVp and a tube current of 10 mA. The effective radiation dose of an ULDCT study was 0.1 mSv. We evaluated the image quality of CT images by using signal-to-noise ratio (SNR) in each image. We applied the supervised dose reduction technique trained with the phantom to the patient cases. With the trained supervised dose reduction technique, noise and artifacts (e.g., streaks) in ULDCT images (0.1 mSv) were reduced substantially, while details of soft tissue such as pulmonary vessels and bones were maintained, as illustrated in FIGS. 4 a/4 b, 5 a/5 b, and 6 a/6 b. In these example implementation, the average SNR for the 0.1 mSv ULDCT images for patients was improved from 2.3 (±1.8) to 13.0 (±2.5) dB (two-tailed t-test; P<.05). This 10.7 dB average SNR improvement was comparable to the 11.5 dB improvement that we were able to achieve by increasing the effective radiation dose from 0.1 mSv (10 mA) to 1.5 mSv (150 mA) in the phantom study, used as a reference, as illustrated by comparing FIGS. 5 a/5 b and 6 a/6 b.)
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAHMINA N ANSARI whose telephone number is (571)270-3379. The examiner can normally be reached on IFP Flex - Monday through Friday 9 to 5.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, O' NEAL MISTRY can be reached on 313-446-4912. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
TAHMINA N. ANSARI
Examiner
Art Unit 2672
2672
May 12, 2026
/TAHMINA N ANSARI/Primary Examiner, Art Unit 2674
Prosecution Insights