Synthesis of medical images of brain tumors using 3D-2D GANs
Systems and methods for generating synthesized medical images of a tumor are provided. A 3D mask of an anatomical structure generated from a 3D medical image and a 3D image of a plurality of concentric spheres are received. A 3D mask of a tumor is generated based on the 3D mask of the anatomical structure and the 3D image of the plurality of concentric spheres using a first 3D generator network. A 3D intensity map of the tumor is generated based on the 3D mask of the tumor and the 3D image of the plurality of concentric spheres using a second 3D generator network. A 3D synthesized medical image of the tumor is generated based on one or more 2D slices of the 3D intensity map of the tumor and one or more 2D slices of the 3D medical image using a 2D generator network. The 3D synthesized medical image of the tumor is output.
1. A computer-implemented method comprising:
receiving 1) a 3D mask of an anatomical structure generated from a 3D medical image and 2) a 3D image of a plurality of concentric spheres;
generating a 3D mask of a tumor based on the 3D mask of the anatomical structure and the 3D image of the plurality of concentric spheres using a first 3D generator network;
generating a 3D intensity map of the tumor based on the 3D mask of the tumor and the 3D image of the plurality of concentric spheres using a second 3D generator network;
generating a 3D synthesized medical image of the tumor based on one or more 2D slices of the 3D intensity map of the tumor and one or more 2D slices of the 3D medical image using a 2D generator network; and
outputting the 3D synthesized medical image of the tumor.
2. The computer-implemented method of claim 1 , further comprising:
smoothing the tumor in the 3D synthesized medical image using a 3D Gaussian kernel;
extracting the smoothed tumor from the 3D synthesized medical image; and
blending the extracted smoothed tumor with the 3D medical image.
3. The computer-implemented method of claim 2 , further comprising:
adjusting a contrast of the extracted smoothed tumor.
4. The computer-implemented method of claim 1 , wherein one or more of the first 3D generator network, the second 3D generator network, or the 2D generator network are trained based on at least one of 3D training images of a plurality of concentric spheres and 3D and 2D training intensity maps of a tumor, the 3D training images of the plurality of concentric spheres and the 3D and 2D training intensity maps of the tumor generated by:
receiving a 3D training image of the tumor and a 3D training mask of the tumor;
applying multi-Otsu thresholding to classify voxels of the 3D training image within the tumor identified in the 3D training mask to generate the 3D training intensity maps of the tumor;
generating the 3D training images of the plurality of concentric spheres based on a volume of the voxels in each of the classes in the 3D training intensity map of the tumor; and
generating the 2D training intensity map of the tumor by unstacking the 3D training intensity map of the tumor along an axis into 2D slices.
5. The computer-implemented method of claim 1 , wherein the 2D generator network is trained with adversarial loss using a 2D discriminator network.
6. The computer-implemented method of claim 1 , wherein the 2D generator network is trained with a 2D perceptual loss.
7. The computer-implemented method of claim 1 , wherein the plurality of concentric spheres comprises three concentric spheres.
8. The computer-implemented method of claim 1 , wherein the anatomical structure is a brain of a patient or a healthy subject.
9. The computer-implemented method of claim 1 , further comprising:
training a machine learning based network for performing a medical imaging analysis task based on the 3D synthesized medical image of the tumor.
10. An apparatus comprising:
means for receiving 1) a 3D mask of an anatomical structure generated from a 3D medical image and 2) a 3D image of a plurality of concentric spheres;
means for generating a 3D mask of a tumor based on the 3D mask of the anatomical structure and the 3D image of the plurality of concentric spheres using a first 3D generator network;
means for generating a 3D intensity map of the tumor based on the 3D mask of the tumor and the 3D image of the plurality of concentric spheres using a second 3D generator network;
means for generating a 3D synthesized medical image of the tumor based on one or more 2D slices of the 3D intensity map of the tumor and one or more 2D slices of the 3D medical image using a 2D generator network; and
means for outputting the 3D synthesized medical image of the tumor.
11. The apparatus of claim 10 , further comprising:
means for smoothing the tumor in the 3D synthesized medical image using a 3D Gaussian kernel;
means for extracting the smoothed tumor from the 3D synthesized medical image; and
means for blending the extracted smoothed tumor with the 3D medical image.
12. The apparatus of claim 11 , further comprising:
means for adjusting a contrast of the extracted smoothed tumor.
13. The apparatus of claim 10 , wherein one or more of the first 3D generator network, the second 3D generator network, or the 2D generator network are trained based on at least one of 3D training images of plurality of a concentric spheres and 3D and 2D training intensity maps of a tumor, the 3D training images of the plurality of concentric spheres and the 3D and 2D training intensity maps of the tumor generated by:
means for receiving a 3D training image of the tumor and a 3D training mask of the tumor;
means for applying multi-Otsu thresholding to classify voxels of the 3D training image within the tumor identified in the 3D training mask to generate the 3D training intensity maps of the tumor;
means for generating the 3D training images of the plurality of concentric spheres based on a volume of the voxels in each of the classes in the 3D training intensity map of the tumor; and
means for generating the 2D training intensity map of the tumor by unstacking the 3D training intensity map of the tumor along an axis into 2D slices.
14. The apparatus of claim 10 , wherein the 2D generator network is trained with adversarial loss using a 2D discriminator network.
15. A non-transitory computer readable medium storing computer program instructions, the computer program instructions when executed by a processor cause the processor to perform operations comprising:
receiving 1) a 3D mask of an anatomical structure generated from a 3D medical image and 2) a 3D image of a plurality of concentric spheres;
generating a 3D mask of a tumor based on the 3D mask of the anatomical structure and the 3D image of the plurality of concentric spheres using a first 3D generator network;
generating a 3D intensity map of the tumor based on the 3D mask of the tumor and the 3D image of the plurality of concentric spheres using a second 3D generator network;
generating a 3D synthesized medical image of the tumor based on one or more 2D slices of the 3D intensity map of the tumor and one or more 2D slices of the 3D medical image using a 2D generator network; and
outputting the 3D synthesized medical image of the tumor.
16. The non-transitory computer readable medium of claim 15 , the operations further comprising:
smoothing the tumor in the 3D synthesized medical image using a 3D Gaussian kernel;
extracting the smoothed tumor from the 3D synthesized medical image; and
blending the extracted smoothed tumor with the 3D medical image.
17. The non-transitory computer readable medium of claim 15 , wherein the 2D generator network is trained with a 2D perceptual loss.
18. The non-transitory computer readable medium of claim 15 , wherein the plurality of concentric spheres comprises three concentric spheres.
19. The non-transitory computer readable medium of claim 15 , wherein the anatomical structure is a brain of a patient or healthy subject.
20. The non-transitory computer readable medium of claim 15 , the operations further comprising:
training a machine learning based network for performing a medical imaging analysis task based on the 3D synthesized medical image of the tumor.