System and method for artifact reduction of computed tomography reconstruction leveraging artificial intelligence and a priori known model for the object of interest
Nondestructive evaluation (NDE) of objects can elucidate impacts of various process parameters and qualification of the object. Computed tomography (CT) enables rapid NDE and characterization of objects. However, CT presents challenges because of artifacts produced by standard reconstruction algorithms. Beam-hardening artifacts especially complicate and adversely impact the process of detecting defects. By leveraging computer-aided design (CAD) models, CT simulations, and a deep-neutral network high-quality CT reconstructions that are affected by noise and beam-hardening can be simulated and used to improve reconstructions. The systems and methods of the present disclosure can significantly improve the reconstruction quality, thereby enabling better detection of defects compared with the state of the art.
1. An artifact reduction artificial intelligence training system for computed tomography (CT) of an object of interest, the system comprising:
a computer-aided design (CAD) model representing the object of interest, stored in memory;
an artifact characterization, stored in memory;
one or more computer subsystems; and
one or more components executed by the one or more computer subsystems, wherein the one or more components include:
a computed tomography (CT) simulator to generate a plurality of CT simulated projections based on the CAD model, wherein a subset of the plurality of CT simulated projections include simulated artifacts based on the artifact characterization, wherein the artifact characterization includes a set of beam-hardening parameters and detector noise parameters for simulating artifacts caused by beam hardening and detector noise during an actual CT scan, wherein the CT simulator is configured to generate at least a subset of the plurality of CT simulated projections based on the CAD model, the beam-hardening parameters, and the detector noise parameters, wherein the generated plurality of corresponding artifact CT simulated projections are realistically noisy, and wherein a subset of the plurality of the CT simulated projections lack simulated artifacts; and
a deep learning component configured to train a deep learning artifact reduction model based on the plurality of CT simulated projections and generate a set of deep learning artifact reduction model parameters.
2. The artifact reduction artificial intelligence training system of claim 1 wherein the deep learning component is configured to reconstruct images of the object of interest based on the CT simulated projections and train the deep learning artifact reduction model in the image domain based on the reconstructed images.
3. The artifact reduction artificial intelligence training system of claim 1 wherein the deep learning component is configured to train the deep learning artifact reduction model in the sinogram domain based directly on the CT simulated projections.
4. The artifact reduction artificial intelligence training system of claim 3 wherein the deep learning component is configured to train the deep learning artifact reduction model without reconstruction of an image of the object of interest.
5. The artifact reduction artificial intelligence training system of claim 1 including a CT imaging system configured to perform a CT scan of a physical version of the object of interest and obtain CT measured projections, wherein the deep learning component is configured to process the CT measured projections to reduce artifacts based on the deep learning artifact reduction model and generate a reconstruction correction, and wherein the one or more components include a reconstruction component configured to reconstruct an image of the object of interest based on the reconstruction correction.
6. The artifact reduction artificial intelligence training system of claim 5 wherein the reconstruction component is configured to reconstruct an image of the object of interest without the reconstruction correction, compare the image reconstructed without the reconstruction correction and the image reconstructed with the reconstruction correction, and output the comparison.
7. The artifact reduction artificial intelligence training system of claim 5 wherein the reconstruction component is configured to reconstruct an image of the object of interest without the reconstruction correction, wherein artifacts in the reconstructed image without the reconstruction correction mask defects of the object of interest in the reconstructed image, and wherein defects of the object of interest are visible in the image reconstructed using the reconstruction correction that reduces the number of artifacts.
8. The artifact reduction artificial intelligence training system of claim 5 wherein the CAD model of the object of interest includes a set of defects positioned to be masked by artifacts created by the CT scan, and wherein the reconstructed image of the object of interest reduces the artifacts such that the set of defects in the object of interest are visible in the reconstructed image.
9. The artifact reduction artificial intelligence training system of claim 1 wherein the one or more components include a realistic defect generator configured to process the CAD model and generate one or more modified versions of the CAD model that represent the object of interest with different sets of simulated realistic defects, and wherein the CT simulator is configured to generate the plurality of CT simulated projections based on the one or more modified versions of the CAD model.
10. The artifact reduction artificial intelligence training system of claim 1 wherein the one or more components include a realistic defect generator configured to train a deep generative model based on the CAD model and one or more training representations derived from physically manufactured versions of the object of interest, wherein the one or more training representations include physical defects, and wherein the realistic defect generator is configured to generate a trained deep generative model with realistic defect model parameters that add realistic defects to representations of the CAD model.
11. The artifact reduction artificial intelligence training system of claim 10 wherein the one or more training representations includes at least one of a high-resolution CT image of the object of interest, a high-resolution scanning electron microscopy image of the object of interest, and a high-resolution transmission electron microscopy image of the object of interest.
12. The artifact reduction artificial intelligence training system of claim 1 wherein the one or more components include a resolution enhancing component configured to train a deep generative model based on the CAD model and one or more training representations derived from physically manufactured versions of the object of interest, wherein the one or more training representations are higher resolution than the CAD model, and wherein the resolution enhancing component is configured to generate a trained deep generative model with enhanced resolution model parameters that increase resolution of features of the CAD model.
13. The artifact reduction artificial intelligence training system of claim 1 wherein the deep learning component is configured to train the deep learning artifact reduction model without measured CT projections from an CT scan of a physical version of the object of interest.
14. The artifact reduction artificial intelligence training system of claim 1 wherein the one or more components include a transfer learning component configured to transform the set of deep learning artifact reduction model parameters trained on CT simulated projections of the object of interest to a set of deep learning artifact reduction model parameters for use on CT measured projections of a physical version of the object of interest.
15. The artifact reduction artificial intelligence training system of claim 1 wherein the artifact characterization includes a set of electronic noise parameters for simulating artifacts caused by electronic noise during an actual CT scan.
16. The artifact reduction artificial intelligence training system of claim 1 wherein the artifact characterization includes a set of CT scan time parameters for simulating artifacts caused by low scan time during an actual CT scan.
17. The artifact reduction artificial intelligence training system of claim 1 wherein the one or more components includes a deep learning deployment component configured to deploy the set of deep learning artifact reduction model parameters to suppress artifacts from reconstructed 3D volumes.
18. The artifact reduction artificial intelligence training system of claim 1 including a reconstruction component configured to reconstruct a 3D volume of the object of interest using a 2.5D scheme by which each slice of the 3D volume is reconstructed from multiple slices of an CT scan to exploit correlations between adjacent slices.
19. An artifact reduction artificial intelligence training method for computed tomography (CT) of an object of interest, the method comprising:
obtaining a computer-aided design (CAD) model representing a 3D volume of the object of interest, stored in memory;
obtaining an artifact characterization, stored in memory;
simulating a plurality of CT simulated projections based on the CAD model, wherein a subset of the plurality of CT simulated projections are artifact CT simulated projections that simulate artifacts based on a combination of the CAD model and the artifact characterization, wherein the artifact characterization includes a set of beam-hardening parameters and a set of detector noise parameters simulating artifacts caused by beam hardening and detector noise during an actual CT scan, wherein simulating a plurality of CT simulated projections comprises generating at least a subset of the plurality of CT simulated projections based on the CAD model, the beam-hardening parameters, and the detector noise parameters, wherein the generated plurality of artifact CT simulated projections are realistically noisy, and wherein a subset of the plurality of the CT simulated projections are non-artifact CT simulated projections that do not include simulated artifacts; and
training a deep learning artifact reduction model based on the subset of the plurality of artifact CT simulated projections and the subset of the plurality of non-artifact CT simulated projections; and generating a set of deep learning artifact reduction model parameters.
20. The artifact reduction artificial intelligence training method of claim 19 including reconstructing images of the 3D volume of the object of interest based on the subset of the plurality of artifact CT simulated projections and the subset of the plurality of non-artifact CT simulated projections, and wherein the training includes training the deep learning artifact reduction model based on the reconstructed images of the 3D volume of the object of interest.
21. The artifact reduction artificial intelligence training method of claim 19 wherein the CT simulated projections are a plurality of sinograms, and the training includes training the deep learning artifact reduction model directly on the plurality of sinograms.
22. The artifact reduction artificial intelligence training method of claim 21 wherein the training includes training the deep learning artifact reduction model without reconstruction of an image of the 3D volume of the object of interest.
23. The artifact reduction artificial intelligence training method of claim 19 including performing a CT scan of a physical version of the object of interest, obtaining CT measured projections of the 3D volume of the physical version of the object of interest, processing the CT measured projections to remove artifacts based on the deep learning artifact reduction model, generating a reconstruction artifact correction, and reconstructing an image of the 3D volume of the physical version of the object of interest based on the reconstruction artifact correction.
24. The artifact reduction artificial intelligence training method of claim 23 wherein the reconstructing includes reconstructing an image of the 3D volume of the physical version of the object of interest without the reconstruction artifact correction and providing a comparison between the reconstructed image of the 3D volume reconstructed without the reconstruction artifact correction and the reconstructed image of the 3D volume reconstructed with the reconstruction artifact correction.
25. The artifact reduction artificial intelligence training method of claim 23 including comparing the reconstructed image of the 3D volume with the reconstruction artifact correction and an image of the 3D volume representing ground truth, and outputting the comparison.
26. The artifact reduction artificial intelligence training method of claim 19 wherein the CAD model includes simulated defects representative of defects caused by additive manufacture of the object of interest.
27. The artifact reduction artificial intelligence training method of claim 19 including generating realistic defects by processing the CAD model and generating one or more modified versions or augmentations of the CAD model that represent the 3D volume of the object of interest with different sets of simulated realistic defects, and generating the plurality of CT simulated projections based on the one or more modified versions or augmentations of the CAD model with different sets of simulated realistic defects.
28. The artifact reduction artificial intelligence training method of claim 19 including generating realistic defects, training a generative-adversarial neural network (GAN) based on one or more training CAD models and one or more training representations derived from physical versions of the object of interest, wherein the one or more training representations include physical manufacturing defects, and generating a trained GAN model with realistic defect model parameters that modify an input CAD model with realistic manufacturing defects, and generating the plurality of CT simulated projections based on the modified input CAD model.
29. The artifact reduction artificial intelligence training method of claim 28 wherein the one or more training representations includes at least one of a high-resolution CT image, a high-resolution scanning electron microscopy image, and a high-resolution transmission electron microscopy image.
30. The artifact reduction artificial intelligence training method of claim 19 including generating realistic defects, training a generative-adversarial neural network (GAN) based on a generative model and a discriminative model, wherein the generative model represents actual XCT measurements of a physical instance of the object of interest and the discriminative model represents the CT simulated projections of the CAD model of the object of interest, and generating a trained GAN model with realistic defect model parameters that modify an input CAD model with realistic manufacturing defects to generate GAN-generated XCT images, and
wherein training the deep learning artifact reduction model includes training the deep learning artifact reduction model based on GAN-generated XCT images.
31. The artifact reduction artificial intelligence training method of claim 19 including training the deep learning artifact reduction model without the use of CT measured projections from a CT scan of a physical version of the object of interest.
32. The artifact reduction artificial intelligence training method of claim 19 including transfer learning the set of deep learning artifact reduction model parameters trained on CT simulated projections to a set of deep learning artifact reduction model parameters for use on CT measured projections of a physical version of the object of interest.
33. The artifact reduction artificial intelligence training method of claim 19 wherein the artifact characterization includes a set of beam-hardening parameters and a set of detector noise parameters.