Methods, systems, and media for discriminating and generating translated images

1. A method for discriminating and generating translated images, comprising:

identifying a set of training images, wherein each image in the set of training images is associated with at least one domain from a plurality of domains, and wherein each domain corresponds to an attribute of an image;

training a generator network of a generative adversarial network to generate, using a first image from the set of training images that is associated with a first domain from the plurality of domains: i) a first fake image that is associated with the first domain from the plurality of domains; and ii) a second fake image that is associated with a second domain from the plurality of domains;

training a discriminator network of the generative adversarial network using as inputs to the discriminator network: i) an image from the set of training images; ii) the first fake image that is associated with the first domain from the plurality of domains; and iii) the second fake image that is associated with the second domain from the plurality of domains; and

in response to determining that training of the generator network and the discriminator network is complete, using the generator network to generate, for an image not included in the set of training images that is associated with the first domain from the plurality of domains at least one of: i) a third fake image that is associated with the first domain from the plurality of domains; and ii) a fourth fake image that is associated with the second domain from the plurality of domains.

2. The method of claim 1 , wherein each image in the set of training images corresponds to an image of a portion of a body of a patient, and wherein each domain in the plurality of domains corresponds to a presence or an absence of a disease state in an image.

3. The method of claim 2 , wherein the first domain corresponds to a presence of a lesion in the image of the portion of the body of the patient, and wherein the second domain corresponds to a lack of a lesion in the image of the portion of the body of the patient.

4. The method of claim 3 , further comprising generating a difference map by subtracting the fourth fake image that is associated with the second domain from the image not included in the training set.

5. The method of claim 1 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated first fake image associated with the first domain of the plurality of domains.

6. The method of claim 1 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated second fake image associated with the second domain of the plurality of domains.

7. A system for discriminating and generating translated images, the system comprising:

a memory; and

a hardware processor that, when executing computer-executable instructions stored in the memory, is configured to:

identify a set of training images, wherein each image in the set of training images is associated with at least one domain from a plurality of domains, and wherein each domain corresponds to an attribute of an image;

train a generator network of a generative adversarial network to generate, using a first image from the set of training images that is associated with a first domain from the plurality of domains: i) a first fake image that is associated with the first domain from the plurality of domains; and ii) a second fake image that is associated with a second domain from the plurality of domains;

train a discriminator network of the generative adversarial network using as inputs to the discriminator network: i) an image from the set of training images; ii) the first fake image that is associated with the first domain from the plurality of domains; and iii) the second fake image that is associated with the second domain from the plurality of domains; and

in response to determining that training of the generator network and the discriminator network is complete, use the generator network to generate, for an image not included in the set of training images that is associated with the first domain from the plurality of domains at least one of: i) a third fake image that is associated with the first domain from the plurality of domains; and ii) a fourth fake image that is associated with the second domain from the plurality of domains.

8. The system of claim 7 , wherein each image in the set of training images corresponds to an image of a portion of a body of a patient, and wherein each domain in the plurality of domains corresponds to a presence or an absence of a disease state in an image.

9. The system of claim 8 , wherein the first domain corresponds to a presence of a lesion in the image of the portion of the body of the patient, and wherein the second domain corresponds to a lack of a lesion in the image of the portion of the body of the patient.

10. The system of claim 9 , wherein the hardware processor is further configured to generate a difference map by subtracting the fourth fake image that is associated with the second domain from the image not included in the training set.

11. The system of claim 7 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated first fake image associated with the first domain of the plurality of domains.

12. The system of claim 7 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated second fake image associated with the second domain of the plurality of domains.

13. A non-transitory computer-readable medium containing computer executable instructions that, when executed by a processor, cause the processor to perform a method for discriminating and generating translated images, the method comprising:

identifying a set of training images, wherein each image in the set of training images is associated with at least one domain from a plurality of domains, and wherein each domain corresponds to an attribute of an image;

training a generator network of a generative adversarial network to generate, using a first image from the set of training images that is associated with a first domain from the plurality of domains: i) a first fake image that is associated with the first domain from the plurality of domains; and ii) a second fake image that is associated with a second domain from the plurality of domains;

training a discriminator network of the generative adversarial network using as inputs to the discriminator network: i) an image from the set of training images; ii) the first fake image that is associated with the first domain from the plurality of domains; and iii) the second fake image that is associated with the second domain from the plurality of domains; and

in response to determining that training of the generator network and the discriminator network is complete, using the generator network to generate, for an image not included in the set of training images that is associated with the first domain from the plurality of domains at least one of: i) a third fake image that is associated with the first domain from the plurality of domains; and ii) a fourth fake image that is associated with the second domain from the plurality of domains.

14. The non-transitory computer-readable medium of claim 13 , wherein each image in the set of training images corresponds to an image of a portion of a body of a patient, and wherein each domain in the plurality of domains corresponds to a presence or an absence of a disease state in an image.

15. The non-transitory computer-readable medium of claim 14 , wherein the first domain corresponds to a presence of a lesion in the image of the portion of the body of the patient, and wherein the second domain corresponds to a lack of a lesion in the image of the portion of the body of the patient.

16. The non-transitory computer-readable medium of claim 15 , wherein the method further comprises generating a difference map by subtracting the fourth fake image that is associated with the second domain from the image not included in the training set.

17. The non-transitory computer-readable medium of claim 13 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated first fake image associated with the first domain of the plurality of domains.

18. The non-transitory computer-readable medium of claim 13 , wherein training the generator network comprises optimizing a generator loss function, and wherein the generator loss function includes a penalty based on the generated second fake image associated with the second domain of the plurality of domains.