Systems and methods for deep learning-based image reconstruction

1. A system comprising:

a regression model trainer to train a regression model, the regression model trained using three-dimensional (3D) volume data and at least one of acquired two-dimensional (2D) projection data or simulated 2D projection data and deployed to reconstruct image pixels to form a reconstructed image;

a pixel identifier to identify a pixel to be reconstructed;

a volume receiver to receive a first volume; and

a pixel reconstructor including:

a pixel mapper to map the pixel onto voxels from the first volume according to an acquisition geometry, the acquisition geometry associated with a particular path or movement of an x-ray source with respect to a 3D object to obtain a series of 2D projections;

an adjacent voxel extractor to extract adjacent voxels around each mapped pixel of the image pixels; and

a regression model feeder to feed the regression model with the extracted adjacent voxels to produce a reconstructed value of the pixel.

2. The system of claim 1 , wherein the regression model trainer includes:

a database including acquired projection data and a 2D mammogram acquired under the same compression, the regression model trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; or

a database including simulated projection data and a simulated 2D mammogram acquired under the same compression from a digital anthropomorphic phantom, the regression model trained to output a 2D image approximately identical to the simulated 2D mammogram when fed with the simulated projections.

3. The system of claim 1 , wherein the regression model trainer includes:

a Digital Anthropomorphic Phantom (DAP) Modeler including an acquisition simulator, an algorithm creator, and a DAP database;

a Computed Tomography (CT) Modeler including an acquisition simulator, an algorithm creator, and a CT database; and

an Algorithm Modifier including an acquisition reconstructor and an algorithm database.

4. The system of claim 1 , further including a feedback generator to identify when a mistake has been made on the reconstructed image and, when the mistake is identified, to communicate to the regression model trainer to re-train the regression model.

5. The system of claim 1 , further including a reconstructed value producer to produce a reconstructed value for each pixel of the image pixels to be reconstructed, the reconstructed values used to produce the reconstructed image.

6. The system of claim 1 , further including a user interface, the user interface to display the reconstructed image pixels.

7. A non-transitory computer readable storage medium comprising instructions which, when executed, cause a processor to at least:

receive a first volume;

identify a pixel to reconstruct;

receive a trained regression model, the regression model trained using three-dimensional (3D) volume data and at least one of acquired two-dimensional (2D) projection data or simulated 2D projection data and deployed to reconstruct image pixels to form a reconstructed image; and

reconstruct the pixel by:

mapping the pixel onto voxels from the first volume according to an acquisition geometry, the acquisition geometry associated with a particular path or movement of an x-ray source with respect to a 3D object to obtain a series of 2D projections;

extracting adjacent voxels around each mapped pixel of the image pixels;

feeding the regression model with the extracted adjacent voxels to produce a reconstructed value of the pixel; and

repeating the reconstruction for each pixel of the image pixels to be reconstructed.

8. The non-transitory computer readable storage medium of claim 7 , wherein the regression model is trained on at least one of:

a database including acquired projection data and a 2D mammogram acquired under the same compression, the regression model trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; or

a database including simulated projection data and a simulated 2D mammogram acquired under the same compression from a digital anthropomorphic phantom, the regression model trained to output a 2D image approximately identical to the simulated 2D mammogram when fed with the simulated projections.

9. The non-transitory computer readable storage medium of claim 7 , wherein the regression model is trained on at least one of:

a database including digital anthropomorphic phantoms and simulated projection data obtained from the phantoms for a given acquisition geometry, the regression model trained to output a second volume approximately identical to the anthropomorphic phantom when fed with the simulated projections;

a database including computed tomography (CT) reconstructed data and simulated projections data obtained from the CT reconstruction data, the regression model trained to output a third volume approximately identical to the CT reconstructed data when fed with the simulated projections; or

a database including acquired projection data and reconstructed data from these projection data with a given reconstruction algorithm, the regression model trained to output a fourth volume approximately identical to the reconstructed data when fed with the acquired projections.

10. The non-transitory computer readable storage medium of claim 7 , further including instructions which, when executed, cause a machine to identify a mistake within the reconstructed image and, when the mistake is identified, to re-train the regression model.

11. The non-transitory computer readable storage medium of claim 7 , further including instructions which, when executed cause a machine to produce the reconstructed image onto a user interface using the reconstructed pixel values.

12. A method comprising:

receiving a set of imaging projection data;

receiving a first volume;

identifying a pixel to reconstruct;

receiving a trained regression model, the regression model trained using three-dimensional (3D) volume data and at least one of acquired two-dimensional (2D) projection data or simulated 2D projection data and deployed to reconstruct image pixels; and

reconstructing the pixel by:

mapping the pixel onto each imaging projection in the set of projections according to an acquisition geometry;

mapping the pixel onto voxels from the first volume according to an acquisition geometry, the acquisition geometry associated with a particular path or movement of an x-ray source with respect to a 3D object to obtain a series of 2D projections;

extracting adjacent pixels around each mapped pixel of the image pixels in the projections;

extracting adjacent voxels around each mapped pixel of the image pixels in the first volume;

feeding the regression model with the extracted adjacent pixels and extracted adjacent voxels to produce a reconstructed value of the pixel; and

repeating the reconstruction for each pixel of the image pixels to be reconstructed.

13. The method of claim 12 , further including training the regression model on at least one of:

a database including acquired projection data and a 2D mammogram acquired under the same compression, the regression model trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; or

a database including simulated projection data and a simulated 2D mammogram acquired under the same compression from a digital anthropomorphic phantom, the regression model trained to output a 2D image approximately identical to the simulated 2D mammogram when fed with the simulated projections.

14. The method of claim 12 , further including training the regression model on at least one of:

a database including digital anthropomorphic phantoms and simulated projection data obtained from the phantoms for a given acquisition geometry, the regression model trained to output a second volume approximately identical to the anthropomorphic phantom when fed with the simulated projections;

a database including computed tomography (CT) reconstructed data and simulated projections data obtained from the CT reconstruction data, the regression model trained to output a third volume approximately identical to the CT reconstructed data when fed with the simulated projections; or

a database including acquired projection data and reconstructed data from these projection data with a given reconstruction algorithm, the regression model trained to output a fourth volume approximately identical to the reconstructed data when fed with the acquired projections.

15. The method of claim 12 , further including displaying a reconstructed image onto a user interface using the reconstructed values.

16. A non-transitory computer readable storage medium comprising instructions which, when executed, cause a processor to at least:

receive a set of imaging projections data;

receive a first volume;

identify a pixel to reconstruct;

receive a trained regression model, the regression model trained using three-dimensional (3D) volume data and at least one of acquired two-dimensional (2D) projection data or simulated 2D projection data and deployed to reconstruct image pixels to form a reconstructed image; and

reconstruct the pixel by:

mapping the pixel onto each imaging projection in the set of imaging projections according to an acquisition geometry, the acquisition geometry associated with a particular path or movement of an x-ray source with respect to a 3D object to obtain a series of 2D projections;

mapping the pixel onto voxels from the first volume according to an acquisition geometry;

extracting adjacent pixels around each mapped pixel of the image pixels in the projections;

extracting adjacent voxels around the mapped pixel of the image pixels in the first volume;

feeding the regression model with the extracted adjacent pixels and the extracted adjacent voxels to produce a reconstructed value of the pixel; and

repeating the reconstruction for each pixel of the image pixels to be reconstructed.

17. The non-transitory computer readable storage medium of claim 16 , wherein the regression model is trained on at least one of:

a database including acquired projection data and a 2D mammogram acquired under the same compression, the regression model trained to output a 2D image approximately identical to the 2D mammogram when fed with the projections; or

a database including simulated projection data and a simulated 2D mammogram acquired under the same compression from a digital anthropomorphic phantom, the regression model trained to output a 2D image approximately identical to the simulated 2D mammogram when fed with the simulated projections.

18. The non-transitory computer readable storage medium of claim 16 , wherein the regression model is trained on at least one of:

a database including digital anthropomorphic phantoms and simulated projection data obtained from the phantoms for a given acquisition geometry, the regression model trained to output a second volume approximately identical to the anthropomorphic phantom when fed with the simulated projections;

a database including computed tomography (CT) reconstructed data and simulated projections data obtained from the CT reconstruction data, the regression model trained to output a third volume approximately identical to the CT reconstructed data when fed with the simulated projections; or

a database including acquired projection data and reconstructed data from these projection data with a given reconstruction algorithm, the regression model trained to output a fourth volume approximately identical to the reconstructed data when fed with the acquired projections.

19. The non-transitory computer readable storage medium of claim 16 , further including instructions which, when executed, cause a machine to identify a mistake within the reconstructed image and to re-train the regression model.

20. The non-transitory computer readable storage medium of claim 16 , further including instructions which, when executed cause a machine to produce the reconstructed image onto a user interface using the reconstructed pixels.