Patent
Atlas literature
Patent
US 11,854,160 B2Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1A illustrates a functional block diagram of a GAN- CIRCLE (generative adversarial network constrained by an identical, residual and cycle learning …
FIG. 2 illustrates a functional block diagram of a genera- tive neural network consistent with several embodiments of the present disclosure;
FIG. 3 illustrates a functional block diagram of a dis- criminative neural network consistent with several embodi- ments of the present disclosure; and
FIG. 4 is a flow chart of GAN-CIRCLE operations according to various embodiments of the present disclosure
FIG. 50 1A.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A system for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the system comprising: a first generative adversarial network (GAN) comprising: a first generative neural network (G) configured to generate an estimated HR image based, at least in part, on a received LR image, and a first discriminative neural network (DY) configured to compare the estimated HR image and a received training HR image; and an optimization module configured to determine an opti-mization function based, at least in part, on the esti-mated HR image, the optimization function containing B₂ at least one loss function, the optimization module further configured to adjust a plurality of neural net-work parameters associated with the first GAN, to optimize the optimization function.
The system of claim 1, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The system of claim 1, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The system of claim 1, wherein the first discriminative neural network (DY) comprises a plurality of discriminator blocks coupled in series, each discriminator block compris-ing a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The system of claim 1, wherein a trained first genera-tive neural network (G) is configured to generate a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A method for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the method comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The method of claim 8, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The method of claim 8, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The method of claim 8, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The method of claim 8, further comprising generating, by a trained first generative neural network (G), a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A computer readable storage device having stored thereon instructions configured to generate a high resolution (HR) computed tomography (CT) image from a low reso-lution (LR) CT image, the instructions that when executed by one or more processors result in the following operations comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and B₂ adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The device of claim 15, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The device of claim 15, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The device of claim 15, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage. ∗ ∗ ∗ ∗ ∗
Layer stacks claimed or described, ordered top of device to substrate.
GAN-CIRCLE CT super-resolution system
No layer stack recorded.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 2
Patent
Atlas literature
Patent
US 11,854,160 B2Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1A illustrates a functional block diagram of a GAN- CIRCLE (generative adversarial network constrained by an identical, residual and cycle learning …
FIG. 2 illustrates a functional block diagram of a genera- tive neural network consistent with several embodiments of the present disclosure;
FIG. 3 illustrates a functional block diagram of a dis- criminative neural network consistent with several embodi- ments of the present disclosure; and
FIG. 4 is a flow chart of GAN-CIRCLE operations according to various embodiments of the present disclosure
FIG. 50 1A.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A system for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the system comprising: a first generative adversarial network (GAN) comprising: a first generative neural network (G) configured to generate an estimated HR image based, at least in part, on a received LR image, and a first discriminative neural network (DY) configured to compare the estimated HR image and a received training HR image; and an optimization module configured to determine an opti-mization function based, at least in part, on the esti-mated HR image, the optimization function containing B₂ at least one loss function, the optimization module further configured to adjust a plurality of neural net-work parameters associated with the first GAN, to optimize the optimization function.
The system of claim 1, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The system of claim 1, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The system of claim 1, wherein the first discriminative neural network (DY) comprises a plurality of discriminator blocks coupled in series, each discriminator block compris-ing a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The system of claim 1, wherein a trained first genera-tive neural network (G) is configured to generate a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A method for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the method comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The method of claim 8, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The method of claim 8, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The method of claim 8, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The method of claim 8, further comprising generating, by a trained first generative neural network (G), a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A computer readable storage device having stored thereon instructions configured to generate a high resolution (HR) computed tomography (CT) image from a low reso-lution (LR) CT image, the instructions that when executed by one or more processors result in the following operations comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and B₂ adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The device of claim 15, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The device of claim 15, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The device of claim 15, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage. ∗ ∗ ∗ ∗ ∗
Layer stacks claimed or described, ordered top of device to substrate.
GAN-CIRCLE CT super-resolution system
No layer stack recorded.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 2
Patent
Atlas literature
Patent
US 11,854,160 B2Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1A illustrates a functional block diagram of a GAN- CIRCLE (generative adversarial network constrained by an identical, residual and cycle learning …
FIG. 2 illustrates a functional block diagram of a genera- tive neural network consistent with several embodiments of the present disclosure;
FIG. 3 illustrates a functional block diagram of a dis- criminative neural network consistent with several embodi- ments of the present disclosure; and
FIG. 4 is a flow chart of GAN-CIRCLE operations according to various embodiments of the present disclosure
FIG. 50 1A.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A system for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the system comprising: a first generative adversarial network (GAN) comprising: a first generative neural network (G) configured to generate an estimated HR image based, at least in part, on a received LR image, and a first discriminative neural network (DY) configured to compare the estimated HR image and a received training HR image; and an optimization module configured to determine an opti-mization function based, at least in part, on the esti-mated HR image, the optimization function containing B₂ at least one loss function, the optimization module further configured to adjust a plurality of neural net-work parameters associated with the first GAN, to optimize the optimization function.
The system of claim 1, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The system of claim 1, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The system of claim 1, wherein the first discriminative neural network (DY) comprises a plurality of discriminator blocks coupled in series, each discriminator block compris-ing a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The system of claim 1, wherein a trained first genera-tive neural network (G) is configured to generate a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A method for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the method comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The method of claim 8, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The method of claim 8, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The method of claim 8, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The method of claim 8, further comprising generating, by a trained first generative neural network (G), a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A computer readable storage device having stored thereon instructions configured to generate a high resolution (HR) computed tomography (CT) image from a low reso-lution (LR) CT image, the instructions that when executed by one or more processors result in the following operations comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and B₂ adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The device of claim 15, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The device of claim 15, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The device of claim 15, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage. ∗ ∗ ∗ ∗ ∗
Layer stacks claimed or described, ordered top of device to substrate.
GAN-CIRCLE CT super-resolution system
No layer stack recorded.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 2
Patent
Atlas literature
Patent
US 11,854,160 B2Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1A illustrates a functional block diagram of a GAN- CIRCLE (generative adversarial network constrained by an identical, residual and cycle learning …
FIG. 2 illustrates a functional block diagram of a genera- tive neural network consistent with several embodiments of the present disclosure;
FIG. 3 illustrates a functional block diagram of a dis- criminative neural network consistent with several embodi- ments of the present disclosure; and
FIG. 4 is a flow chart of GAN-CIRCLE operations according to various embodiments of the present disclosure
FIG. 50 1A.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A system for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the system comprising: a first generative adversarial network (GAN) comprising: a first generative neural network (G) configured to generate an estimated HR image based, at least in part, on a received LR image, and a first discriminative neural network (DY) configured to compare the estimated HR image and a received training HR image; and an optimization module configured to determine an opti-mization function based, at least in part, on the esti-mated HR image, the optimization function containing B₂ at least one loss function, the optimization module further configured to adjust a plurality of neural net-work parameters associated with the first GAN, to optimize the optimization function.
The system of claim 1, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The system of claim 1, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The system of claim 1, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The system of claim 1, wherein the first discriminative neural network (DY) comprises a plurality of discriminator blocks coupled in series, each discriminator block compris-ing a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The system of claim 1, wherein a trained first genera-tive neural network (G) is configured to generate a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A method for generating a high resolution (HR) com-puted tomography (CT) image from a low resolution (LR) CT image, the method comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The method of claim 8, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The method of claim 8, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The method of claim 8, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The method of claim 8, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage.
The method of claim 8, further comprising generating, by a trained first generative neural network (G), a HR CT image from an actual LR CT image, the actual LR CT image corresponding to low dose CT image data.
A computer readable storage device having stored thereon instructions configured to generate a high resolution (HR) computed tomography (CT) image from a low reso-lution (LR) CT image, the instructions that when executed by one or more processors result in the following operations comprising: generating, by a first generative neural network (G), an estimated HR image based, at least in part, on a received LR image; comparing, by a first discriminative neural network (DY), the estimated HR image and a received training HR image, the first generative neural network (G) and the first discriminative neural network (DY) included in a first generative adversarial network (GAN); determining, by an optimization module, an optimization function based, at least in part, on the estimated HR image, the optimization function containing at least one loss function; and B₂ adjusting, by the optimization module, a plurality of neural network parameters associated with the first GAN, to optimize the optimization function.
The device of claim 15, wherein the at least one loss function is selected from the group comprising an adver-sarial loss function configured to implement a Wasserstein distance with gradient penalty, a cyclic loss function con-figured to implement a cycle consistency constraint, an identity loss function and a joint sparsifying transform loss function.
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function and the received LR image corresponds to a reconstructed LR image received from a second generative neural network (F), F included in a second GAN, the second GAN further comprising a second discriminative neural network (DX).
The device of claim 15, wherein the at least one loss function comprises a cyclic loss function configured to implement an L norm.
The device of claim 15, wherein the first generative neural network (G) comprises a feature extraction network and a reconstruction network, the feature extraction network comprising a plurality of skip connections.
The device of claim 15, wherein the first discrimina-tive neural network (DY) comprises a plurality of discrimi-nator blocks coupled in series, each discriminator block comprising a convolutional stage, a bias stage, an instance norm stage, and a leaky rectified linear unit (ReLU) stage. ∗ ∗ ∗ ∗ ∗
Layer stacks claimed or described, ordered top of device to substrate.
GAN-CIRCLE CT super-resolution system
No layer stack recorded.
Patents and literature cited by this patent (applicant and examiner references).
Cited patents · 2
