Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statements (IDS) were filed on 05/21/2024 and 09/05/2025. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
Applicant first argues the 112(f) interpretation by mentioning the amendment applied to claim 17, examiner agrees and removes the 112(f) interpretation. Applicant then argues the amended independent claim 1 by incorporating the now cancelled claim 6 by mentioning that cited prior art Bacher does not teach the claim limitations of the amended claim. It is also mentioned that cited prior art Li does not cure the deficiencies in Bacher and the latent space in Li does not relate to degrees of freedom of movement to be used in the projection matrix alongside the latent space, examiner disagrees and shows reasoning. Reference Bacher, as seen in the claim rejection of independent claim 1, teaches a further training projection matrix based projection maps and the initial training projection matrix alongside motion parameters of the projections (information relating to dynamic degrees of freedom). The claim limitation “acquiring information relating to dynamic degrees of freedom of movement” has a broad BRI (broadest reasonable interpretation) since it merely acquires information relating to movement in order to obtain a latent space, and one of ordinary skill in the art would interpret it as such. Since Bacher does not specifically teach the use of a “latent space” (any latent space within a BRI), Li comes in to cure the deficiencies and provides clear motivation under the KSR rationales (See MPEP 2143). By utilizing the latent space presented in Li, there is a clear improvement since it simplifies the variables (parameters of Bacher) while retaining variability (for differences in processes) and it is used for reducing computing power and time for the high number of parameters. Therefore one of ordinary skill in the art would come to the conclusion of including the latent space for the further projection matrix for better management of a high number of parameters (merely information relating to movement). Therefore, all of the claims stay rejected and Li is maintained for the rejection of the amended independent claims.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-3, 5, 7-9, 11-13 and 15-19 are rejected under 35 U.S.C. 103 as being unpatentable over Bacher et. al. (Bacher, V., et al. "Learning projection matrices for marker free motion compensation in weight-bearing CT scans." Proc. Fully 3D. Vol. 16. 2021.) in view of Li et. al. (US Pub. No. 20230349277 A1) .
As per claim 1, Baches teaches “A computer-implemented method for providing a three-dimensional (3D) results data set, the method comprising:
acquiring projection maps of an object under examination that are captured from various projection directions by a medical X-ray device;” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” Bacher)
“providing an initial projection matrix based on a static model of the medical X-ray device;” (See fig. 1 (It shows the initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” Bacher)
“acquiring information relating to dynamic degrees of freedom of movement of the medical X-ray device…” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” In addition, the following are also interpreted as projection directions, see page 2 column 2 paragraph 2-3 “The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections… ” See also page 3 figs. 2-3 and section 5 Discussion, fig. 2-3 shows dynamic degrees of movement. See also fig. 4 which shows the final 3d data set. See also section 2 Definition of the Model. See also equations 1 and 2 used for the various projections. Therefore it falls within a BRI (broadest reasonable interpretation) of “dynamic degrees of movement of freedom”. See also pages 2-3 section Results, which shows the angles and dynamic degree movement. Bacher)
“providing a further projection matrix by applying a trained function to input data, wherein the input data is based on the initial projection matrix, the projection map and the… space, wherein at least one parameter of the trained function is adapted based on an image quality metric and/or a consistency metric, and wherein the further projection matrix is provided as output data of the trained function; and” (See fig. 1 (It shows the initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. Fig. 1 also shows the use of equations 1 and 2 which include the trainable weights (A(wi) and A^T (wi) ) applied to the input data. At least one parameter is based on an image quality metric since they are used for motion compensation. Page 2 column 1 shows “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation.” Section 3 paragraph 1-2 shows “The experiments aim to evaluate the possibility of iteratively reducing motion artifacts in cone-beam CT data using a PYRO-NN based trainable reconstruction pipeline. The first experiment investigates the possibility to achieve meaningful gradients for the layer weights. For this, the model shown in Fig. 1 is further simplified to just do a reconstruction and compute a loss with a motion-free label reconstruction. The second experiment is conducted with the presented model to be free of supervision and is inspired by iterative reconstruction methods… The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections. ” The trainable parameters are for reducing motion artifacts (therefore within a Broadest Reasonable Interpretation of a quality metric) as seen on pages 2-3 section Results and along with figs. 1-3 and equations 1-2. In addition, it also teaches within a BRI a consistency metric such as variance, Loss and offset. Also on page 3 section 5 Discussion “… it is possible to interpret the geometric parameters of the CT operators as trainable layers and successfully update these layers weight in a training setting. With the learned parameters the image quality of the tomographic can be improved…”, in addition it also alludes to the use of sharpness metric (another quality based metric). Bacher)
“providing the 3D results data set through reconstruction from the projection maps by the further projection matrix.” (See fig. 1, and the provided 3d results data set is shown on page 4 fig. 4. See also pages 2-3 sections 3 Experiments Description and section 4 Results along with section 5. They all discuss the resulting data set. See also all of page 1-2 section Definition of the Model and page 2 column 2 paragraph 2 which shows “The data used in the experiments are simulated 3D Shepp-Logan phantoms. The phantom measures 60 X 60 X 60 mm³ with a pixel size of 0.5 0.5 0.5 Rigid motion is modeled by uniformly random shifts and rotations in the range of to 2 [mm] or [°], respectively, for each projection with which the Shepp-Logan phantom is forward projected to obtain the motion-corrupted sinogram to reconstruct.” Bacher)
, however Bacher does not teach “therein defining a latent space, and wherein the input data based on… latent space.”
Li teaches “therein defining a latent space, and wherein the input data based on… latent space.” (See paragraphs 79-81 “[0079] In some implementations, a synthetic subsurface representation within the layer space may be generated by a trained machine learning model based on input of a latent space vector to the trained machine learning model. The latent space vector may include a sequence of numbers. The value of latent space vector may control the how the synthetic subsurface representation is generated by the trained machine learning model.”)
It would have been obvious to one of ordinary skill in the art before the effective filing
date of the claimed invention to combine the teachings of Bacher with the teachings of Li to utilize a latent space as input. The modification would have been motivated by the desire to simplify the variables while retaining variability to reduce the needed computing power (improves performance), therefore it is an improvement, as suggested by Li (“[0080] The latent space vector may serve as a low dimensional parameterization of subsurface representation generation. Traditional modeling of subsurface representation may include a high number of parameters, and identifying the right combination of parameter values to generate subsurface representations that match conditioning characteristics may be difficult and costly (e.g., in terms of computing power and/or time). The latent space vector may simplify the variables used to generate subsurface representation while retaining the ability to achieve same variability in the subsurface representations.” Li)
Claim 17 is rejected under the same analysis as claim 1. (Li mentions memory and processors (See paragraph 93.) Bacher mentions the use of a computations for use of image reconstruction along the use of neural networks for use in computers, it is well known in the art that computers utilize processor and memory for processing data (See abstract and Introduction on page 1).)
Claim 18 is rejected under the same analysis as claim 1. “(See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” Bacher)
As per claim 2, Bacher in view of Li teaches “The method of claim 1, wherein an initial projection matrix and a further projection matrix are provided for each projection direction of the various projection directions, wherein the input data of the trained function is based on the initial projection matrices, and wherein the 3D results data set is provided through reconstruction from the projection maps by the further projection matrices.” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” In addition, the following are also interpreted as projection directions, see page 2 column 2 paragraph 2-3 “The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections… ” See also page 3 figs. 2-3 and section 5 Discussion. See also fig. 4 which shows the final 3d data set. See also section 2 Definition of the Model. See also equations 1 and 2 used for the various projections. Bacher)
As per claim 3, Bacher in view of Li teaches “The method of claim 2, wherein the input data of the trained function is additionally based on the static model of the medical X-ray device.” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. See also sections 4 and 5. Bacher)
As per claim 5, Bacher in view of Li teaches “the method of claim 1, wherein the input data of the trained function is additionally based on the static model of the medical X-ray device.” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. Bacher)
As per claim 7, Bacher teaches “A computer-implemented method for providing a trained function, the method comprising:
acquiring training projection maps of a training object under examination which map the training object under examination from various projection directions, wherein the training projection maps are simulated or captured by a medical training X-ray device;” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” Bacher)
“providing an initial training projection matrix based on a static training model of the medical training X-ray device;” (See fig. 1 (It shows the initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” Bacher)
“acquiring information relating to dynamic degrees of freedom of movement of the medical X-ray device… defining a… training space” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” In addition, the following are also interpreted as projection directions, see page 2 column 2 paragraph 2-3 “The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections… ” See also page 3 figs. 2-3 and section 5 Discussion, fig. 2-3 shows dynamic degrees of movement. See also fig. 4 which shows the final 3d data set. See also section 2 Definition of the Model. See also equations 1 and 2 used for the various projections. Therefore it falls within a BRI (broadest reasonable interpretation) of “dynamic degrees of movement of freedom”. See also pages 2-3 section Results, which shows the angles and dynamic degree movement. Bacher)
“providing a further training projection matrix by applying the trained function to input data, wherein the input data is based on the training projection maps, the initial training projection matrix and… training space, and wherein the further training projection matrix is provided as output data of the trained function;” (See fig. 1 (It shows the initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. Fig. 1 also shows the use of equations 1 and 2 which include the trainable weights (A(wi) and A^T (wi) ) applied to the input data. At least one parameter is based on an image quality metric since they are used for motion compensation. Page 2 column 1 shows “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation.” Section 3 paragraph 1-2 shows “The experiments aim to evaluate the possibility of iteratively reducing motion artifacts in cone-beam CT data using a PYRO-NN based trainable reconstruction pipeline. The first experiment investigates the possibility to achieve meaningful gradients for the layer weights. For this, the model shown in Fig. 1 is further simplified to just do a reconstruction and compute a loss with a motion-free label reconstruction. The second experiment is conducted with the presented model to be free of supervision and is inspired by iterative reconstruction methods… The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections. ” See also all of page 1-2 section Definition of the Model and page 2 column 2 paragraph 2 which shows “The data used in the experiments are simulated 3D Shepp-Logan phantoms. The phantom measures 60 X 60 X 60 mm³ with a pixel size of 0.5 0.5 0.5 Rigid motion is modeled by uniformly random shifts and rotations in the range of to 2 [mm] or [°], respectively, for each projection with which the Shepp-Logan phantom is forward projected to obtain the motion-corrupted sinogram to reconstruct. Therefore also used as output. Bacher)
“reconstructing a three-dimensional (3D) training data set from the training projection maps by the further training projection matrix;” (See fig. 1, and the provided 3d results data set is shown on page 4 fig. 4. See also pages 2-3 sections 3 Experiments Description and section 4 Results along with section 5. They all discuss the resulting data set. See also all of page 1-2 section Definition of the Model and page 2 column 2 paragraph 2 which shows “The data used in the experiments are simulated 3D Shepp-Logan phantoms. The phantom measures 60 X 60 X 60 mm³ with a pixel size of 0.5 0.5 0.5 Rigid motion is modeled by uniformly random shifts and rotations in the range of to 2 [mm] or [°], respectively, for each projection with which the Shepp-Logan phantom is forward projected to obtain the motion-corrupted sinogram to reconstruct.” Bacher)
“determining an evaluation parameter by applying an image quality metric and/or a consistency metric to the 3D-training data set;” (See fig. 1 (It shows the initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. Fig. 1 also shows the use of equations 1 and 2 which include the trainable weights (A(wi) and A^T (wi) ) applied to the input data. At least one parameter is based on an image quality metric since they are used for motion compensation. Page 2 column 1 shows “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation.” Section 3 paragraph 1-2 shows “The experiments aim to evaluate the possibility of iteratively reducing motion artifacts in cone-beam CT data using a PYRO-NN based trainable reconstruction pipeline. The first experiment investigates the possibility to achieve meaningful gradients for the layer weights. For this, the model shown in Fig. 1 is further simplified to just do a reconstruction and compute a loss with a motion-free label reconstruction. The second experiment is conducted with the presented model to be free of supervision and is inspired by iterative reconstruction methods… The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections. ” The trainable parameters are for reducing motion artifacts (therefore within a Broadest Reasonable Interpretation of a quality metric) as seen on pages 2-3 section Results and along with figs. 1-3 and equations 1-2. In addition, it also teaches within a BRI a consistency metric such as variance, Loss and offset. Also on page 3 section 5 Discussion “… it is possible to interpret the geometric parameters of the CT operators as trainable layers and successfully update these layers weight in a training setting. With the learned parameters the image quality of the tomographic can be improved…”, in addition it also alludes to the use of sharpness metric (another quality based metric). Bacher)
“adapting at least one parameter of the trained function based on a comparison of the evaluation parameter with a reference value; and providing the trained function.” (See pages 2-3 section 4 Results “In Fig. 2a the variance of the difference over all 30 projections of the layer weights with respect to the ground truth parameter over the training procedure are plotted. For all parameters, except Φ,, which describes the axis of the scanner's rotation, the ground truth parameter can be learned. Fig. 2c indicates that the high variance is due to a point symmetrical error of the rotation angle. The loss (cf. Fig. 2b) correlates well with the variance of the shifts, while the rotations do not have a large effect on the loss. Fig. 3 shows the results of the second experiment. While for the shifts the true parameter could be found, this is not true for the rotations. Fig. 3c plots the difference between the learned parameters and the ideal ones. The shifts approach nearly zero, with a small negative constant offset for ty. The rotations show random behaviour. Fig. 4 shows multi-planar views of the reconstruction of the phantom before and after the motion correction” It shows that the trainable parameters are adapted based on a comparison to the ground truth (reference value) in order to train. See also figs. 2-3 on page 3. See also sections 2 Definition of the Model and section 3 Experiments discussion (these sections show the trainable parameters and provided training function) and section 5 Discussion. See also equations 1 and 2. Bacher ) , however Bacher does not teach “therein defining a latent space, and wherein the input data based on… latent space.”
Li teaches “therein defining a latent space, and wherein the input data based on… latent space.” (See paragraphs 79-81 “[0079] In some implementations, a synthetic subsurface representation within the layer space may be generated by a trained machine learning model based on input of a latent space vector to the trained machine learning model. The latent space vector may include a sequence of numbers. The value of latent space vector may control the how the synthetic subsurface representation is generated by the trained machine learning model.”)
It would have been obvious to one of ordinary skill in the art before the effective filing
date of the claimed invention to combine the teachings of Bacher with the teachings of Li to utilize a latent space as input. The modification would have been motivated by the desire to simplify the variables while retaining variability to reduce the needed computing power (improves performance), therefore it is an improvement, as suggested by Li (“[0080] The latent space vector may serve as a low dimensional parameterization of subsurface representation generation. Traditional modeling of subsurface representation may include a high number of parameters, and identifying the right combination of parameter values to generate subsurface representations that match conditioning characteristics may be difficult and costly (e.g., in terms of computing power and/or time). The latent space vector may simplify the variables used to generate subsurface representation while retaining the ability to achieve same variability in the subsurface representations.” Li)
Claim 19 is rejected under the same analysis as claim 7. (Li mentions memory and processors (See paragraph 93.) Bacher mentions the use of a computations for use of image reconstruction along the use of neural networks for use in computers, it is well known in the art that computers utilize processor and memory for processing data (See abstract and Introduction on page 1).)
As per claim 8, Bacher in view of Li teaches “The method of claim 7, wherein a 3D comparison data set is reconstructed from the training projection maps by the initial training projection matrix, and wherein the reference value is determined by applying the image quality metric and/or the consistency metric to the 3D comparison data set.” (See page 2 column 1 “Consequently our model for motion compensation takes a motion-corrupted sinogram, which is reconstructed by the first layer. The output of the first layer is directly forward projected using the same geometric parameters as the first layer. The resulting re-projected sinogram is evaluated by a loss against the original sinogram.” A 3d reconstruction (such as that shown in fig. 4 and fig. 1) is performed based on the reference value as seen on page 2 column 2 and section 4 Results “The results of the first experiment are shown in Figure 2 and show the training results for a single scan of a randomly initialized phantom, where these results are representative of other randomly initialized phantoms and motion. In Fig. 2a the variance of the difference over all 30 projections of the layer weights with respect to the ground truth parameter over the training procedure are plotted. For all parameters, except Φ,, which describes the axis of the scanner's rotation, the ground truth parameter can be learned. Fig. 2c indicates that the high variance is due to a point symmetrical error of the rotation angle… Fig. 3 shows the results of the second experiment. While for the shifts the true parameter could be found, this is not true for the rotations. Fig. 3c plots the difference between the learned parameters and the ideal ones… Fig. 4 shows multi-planar views of the reconstruction of the phantom before and after the motion correction…” See also page 3 section 5 Discussion. Bacher )
As per claim 9, Bacher in view of Li teaches “The method of claim 8, wherein an initial training projection matrix and a further training projection matrix are provided for each projection direction of the various projection directions, wherein the input data of the trained function is based on the initial training projection matrices, and wherein the 3D training data set is provided through reconstruction from the training projection maps by the further training projection matrices.” (See abstract and page 2 column 1 and fig.1. Figure 1 shows the 2d projection maps of an object obtained (the sinogram). As seen, there is an x and y sinogram (various projection directions) and column 1 paragraph 1 says “To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non static description of the extrinsic parameters consisting of three rotations and three translations. With this, we define the x-axis parallel to the detector plane horizontal, the scanner rotation is about the y-axis, and the z-axis points towards the detector.” In addition, the following are also interpreted as projection directions, see page 2 column 2 paragraph 2-3 “The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections… ” See also page 3 figs. 2-3 and section 5 Discussion. See also fig. 4 which shows the final 3d data set. See also section 2 Definition of the Model. See also equations 1 and 2 used for the various projections. Bacher)
As per claim 11, Bacher in view of Li already teaches “the method of claim 9, wherein the training projection maps for the various projection directions are simulated based on the static training model” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. Section 3 paragraph 1-2 shows “The experiments aim to evaluate the possibility of iteratively reducing motion artifacts in cone-beam CT data using a PYRO-NN based trainable reconstruction pipeline. The first experiment investigates the possibility to achieve meaningful gradients for the layer weights. For this, the model shown in Fig. 1 is further simplified to just do a reconstruction and compute a loss with a motion-free label reconstruction. The second experiment is conducted with the presented model to be free of supervision and is inspired by iterative reconstruction methods… The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections. ”Bacher) and Li already teaches “and the latent training space.” The same motivation of claim 7 applies here too.
As per claim 13, Bacher in view of Li teaches “The method of claim 7, wherein the input data of the trained function is additionally based on the static training model of the medical training X-ray device.” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. See also sections 4 and 5. Bacher)
As per claim 15, Bacher in view of Li already teaches “The method of claim 7, wherein the training projection maps for the various projection directions are simulated based on the static training model” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. Section 3 paragraph 1-2 shows “The experiments aim to evaluate the possibility of iteratively reducing motion artifacts in cone-beam CT data using a PYRO-NN based trainable reconstruction pipeline. The first experiment investigates the possibility to achieve meaningful gradients for the layer weights. For this, the model shown in Fig. 1 is further simplified to just do a reconstruction and compute a loss with a motion-free label reconstruction. The second experiment is conducted with the presented model to be free of supervision and is inspired by iterative reconstruction methods… The trainable variables consisted of the five motion parameters Wᵢ… for each of the 30 projections. ”Bacher) and Li already teaches “and the latent training space.” The same motivation of claim 7 applies here too.
As per claim 16, Bacher in view of Li already teaches “The method of claim 15, wherein the input data of the trained function is additionally based on the static training model of the medical training X-ray device.” (See fig. 1 (It shows a model that includes initial and further projection matrix) and pages 1 section 2 Definition of the model and page 2 columns 1-2 sections 2-3. “Trainable layers allow setups to update the geometric situation of the projection process and therefore, allow to setup pipelines for motion compensation. A naive implementation where we learn the projections matrices directly would lead to N X 3 4 trainable weights for N projections in the trajectory. To reduce the amount of trainable parameters we used a static definition of the intrinsic camera parameters and a non-static description of the extrinsic parameters consisting of three rotations and three translations.” See also equations 1-2. See also sections 4 and 5. Bacher)
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Bacher in view of Li and further in view of Michael Manhart, hereafter Manhart (US Pub. No. 20210056735 A1 ) .
As per claim 12, Bacher in view of Li already teaches “The method of claim 7,” and “and/or wherein the image quality metric evaluates a total variation of the 3D training data set.” (See page 3 section 5 Discussion “This shortcoming could be overcome by introducing a weighted loss, focusing more on the rotation parameters or by choosing a composite loss function including metrics more sensitive to the artifacts introduced by the rotation parameters, e.g., incorporating sharpness metrics of the reconstructed image such as total variation. In 2017, Sisniega et al. published promising results using such a metric.” Bacher, therefore it teaches the BRI of the whole claim) , Bacher also already teaches “3D training data set” however, Bacher does not teach “wherein the consistency metric evaluates an epipolar consistency and/or a consistency of forward projections of the 3D training data set with the training projection maps, and/or wherein the image quality metric evaluates a total variation of the 3D training data set.”
Manhart teaches “wherein the consistency metric evaluates an epipolar consistency and/or a consistency of forward projections of the 3D training data set with the training projection maps,” (See paragraph 42 “[0042] The consistency function may, for example, be differentiated (e.g., constantly) in accordance with the reconstruction parameters. The consistency function may be embodied to map the reconstruction parameters and the medical measuring data onto at least one scalar (e.g., the at least one consistency value) in order to evaluate a consistency between the reconstruction parameters and the medical image data. The evaluation of the consistency (e.g., with respect to a consistency condition) may, for example, include an epipolar consistency condition (e.g., in accordance with the Grangeat theorem) and/or a sampling consistency condition (e.g., in accordance with the Nyquist sampling theorem) and/or a symmetry consistency condition. The consistency between the reconstructed medical image data and the received medical measuring data may be determined, for example, by forward projection of the reconstructed medical image data.” Manhart) “and/or wherein the image quality metric evaluates a total variation of the 3D training data set.” (See paragraph 31 “[0031] The image quality metric may include, for example, a regression of a back projection error (e.g., with a movement correction). The image quality metric may be embodied to evaluate the image quality with respect to an x-ray beam hardening and/or a signal-to-noise ratio and/or a specification of image artifacts (e.g., metal artifacts). The image quality metric may include an entropy of a gray-scale value histogram and/or a total variation of the image values.” )
It would have been obvious to one of ordinary skill in the art before the effective filing
date of the claimed invention to combine the teachings of Bacher with the teachings of Manhart to utilize an epipolar consistency or consistency. The modification would have been motivated by the desire to have higher image quality, therefore it is an improvement, as suggested by Manhart (See paragraph 3 and 4 “[0003] A high image quality is essential for the analysis and evaluation of medical image data of an examination object. [0004] Medical measuring data (e.g., raw data) of the examination object is frequently recorded by a medical imaging device (e.g., a magnetic resonance system and/or an x-ray device and/or a computed tomography system). A reconstruction of the medical image data from the medical measuring data may then take place. High consistency and precise knowledge of the recording parameters (e.g., a recording geometry) is to be provided for the artifact-free reconstruction of the medical image data.” See also paragraphs 40-49 “[0042] The consistency function may, for example, be differentiated (e.g., constantly) in accordance with the reconstruction parameters. The consistency function may be embodied to map the reconstruction parameters and the medical measuring data onto at least one scalar (e.g., the at least one consistency value) in order to evaluate a consistency between the reconstruction parameters and the medical image data. ” “[0045] In this way, the consistency function may be taken into account as an auxiliary condition of the minimization of the cost value.” See also paragraph 40 “[0040] In this way, a particularly reliable and comprehensive evaluation of the image quality of the reconstructed medical image data may be enabled by applying the cost function. As a result of the cost function being based on a trained function, an evaluation of the image quality of the reconstructed medical image data may be enabled with respect to a number of image features. For example, the cost function (e.g., the trained function) may be based on a number of image quality metrics. The cost function may therefore be embodied, for example, to evaluate the image quality of the reconstructed medical image data with respect to an x-ray beam hardening and/or a signal-to-noise ratio and/or a characteristic of image artifacts (e.g., metal artifacts).” Manhart)
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 DYLAN J MENDEZ MUNIZ whose telephone number is (703)756-5672. The examiner can normally be reached M-F, 8AM - 5PM ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Vu Le can be reached at (571) 272-7332. 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.
/DYLAN JOHN MENDEZ MUNIZ/Examiner, Art Unit 2675
/VU LE/Supervisory Patent Examiner, Art Unit 2668
Prosecution Insights