Training a CNN with pseudo ground truth for CT artifact reduction
Training a CNN with pseudo ground truth for CT artifact reduction is described. An estimated ground truth apparatus is configured to generate an estimated ground truth image based, at least in part, on an initial CT image that includes an artifact. Feature addition circuitry is configured to add a respective feature to each of a number, N, copies of the estimated ground truth image to create the number, N, initial training images. A computed tomography (CT) simulation circuitry is configured to generate a plurality of simulated training CT images based, at least in part, on at least some of the N initial training images. An artifact reduction circuitry is configured to generate a plurality of input training CT images based, at least in part, on the simulated training CT images. A CNN training circuitry is configured to train the CNN based, at least in part, on the input training CT images and based, at least in part, on the initial training images.
1. A method for computed tomography (CT) artifact reduction, the method comprising:
generating, by an estimated ground truth apparatus, an estimated ground truth image based, at least in part, on an initial CT image, the initial CT image comprising an artifact;
adding, by feature addition circuitry, a respective feature to each of a number, N, copies of the estimated ground truth image to create the number, N, initial training images;
generating, by a CT simulation circuitry, a plurality of simulated training CT images based, at least in part, on at least some of the N initial training images, each of at least some of the plurality of simulated training CT images containing at least one respective simulated artifact;
generating, by artifact reduction circuitry, a plurality of input training CT images based, at least in part, on the simulated training CT images; and
training, by a convolutional neural network (CNN) training circuitry, a CNN based, at least in part, on the input training CT images and based, at least in part, on the initial training images.
2. The method of claim 1 , wherein generating the estimated ground truth image comprises:
generating, by the artifact reduction circuitry, an intermediate CT image based, at least in part, on the initial CT image, the intermediate CT image comprising a reduced artifact,
the estimated ground truth image generated based, at least in part, on the intermediate CT image.
3. The method of claim 1 , further comprising extracting, by the feature addition circuitry, a number, M, training patches from each of at least some of the N initial training images, the simulated training CT images generated based, at least in part, on at least some of the M training patches.
4. The method of claim 1 , further comprising validating, by the CNN training circuitry, the trained CNN based, at least in part, on the intermediate CT image.
5. The method of claim 1 , further comprising reducing, by the trained CNN, an actual artifact in an actual CT image.
6. The method of claim 1 , wherein the artifact reduction circuitry corresponds to metal artifact reduction circuitry.
7. The method of claim 1 , wherein the initial CT image is an output of a CT scanner configured to image an imaging object, the imaging object containing a metal implant.
8. A convolutional neural network (CNN) system comprising:
an estimated ground truth apparatus configured to generate an estimated ground truth image based, at least in part, on an initial CT image, the initial CT image comprising an artifact;
feature addition circuitry configured to add a respective feature to each of a number, N, copies of the estimated ground truth image to create the number, N, initial training images;
a first computed tomography (CT) simulation circuitry configured to generate a plurality of simulated training CT images based, at least in part, on at least some of the N initial training images, each of at least some of the plurality of simulated training CT images containing at least one respective simulated artifact;
a first artifact reduction circuitry configured to generate a plurality of input training CT images based, at least in part, on the simulated training CT images; and
a CNN training circuitry configured to train the CNN based, at least in part, on the input training CT images and based, at least in part, on the initial training images.
9. The system of claim 8 , wherein the estimated ground truth apparatus comprises:
a second artifact reduction circuitry configured to generate an intermediate CT image based, at least in part, on the initial CT image, the intermediate CT image comprising a reduced artifact,
the estimated ground truth image generated based, at least in part, on the intermediate CT image.
10. The system of claim 9 , wherein at least one of the first and second artifact reduction circuitry is metal artifact reduction circuitry configured to perform metal artifact reduction.
11. The system of claim 8 , wherein the feature addition circuitry is configured to extract a number, M, training patches from each of at least some of the N initial training images, the simulated training CT images generated based, at least in part, on at least some of the M training patches.
12. The system of claim 8 , wherein the CNN training circuitry is configured to validate the trained CNN based, at least in part, on the intermediate CT image.
13. The system of claim 8 , wherein the trained CNN is configured to reduce an actual artifact in an actual CT image.
14. The system of claim 8 , wherein the initial CT image is an output of a CT scanner configured to image an imaging object, the imaging object containing a metal implant.
15. The system of claim 8 , wherein the initial CT image is simulated based, at least in part, on a phantom image.