Iterative image reconstruction framework
AssigneesGeneral Electric Company, Rensselaer Polytechnic Institute
InventorsLin Fu, Sathish Ramani, Jie Tang, Bruno Kristiaan Bernard De Man, Jed Douglas Pack, Jiang Hsieh, Ge Wang
Life of this patent
Estimating the term…
Abstract
The present disclosure relates to image reconstruction with favorable properties in terms of noise reduction, spatial resolution, detail preservation and computational complexity. The disclosed techniques may include some or all of: a first-pass reconstruction, a simplified datafit term, and/or a deep learning denoiser. In various implementations, the disclosed technique is portable to different CT platforms, such as by incorporating a first-pass reconstruction step.
Claims12
A method for reconstructing an image, comprising:
- receiving a sinogram input;
- performing a first denoising operation on the sinogram input without reconstructing the sinogram input using a first trained artificial neural network;
- generating one or multiple intermediate sinogram(s) from the sinogram input or from one or multiple intermediate image(s) generated from the sinogram input;
- iteratively processing the intermediate sinogram(s), wherein the iterative processing comprises performing at least an iterative datafit operation comparing an output of an iterative loop with an input to the iterative loop;
- performing a second denoising operation on the intermediate sinogram(s) in addition to the datafit operation, wherein the second denoising operation is performed using a second trained artificial neural network, and wherein the datafit operation and the second denoising operation are combined within the iterative loop;
- and
- upon completion of the iterative processing, reconstructing a final image and outputting the final image for review, storage, or further processing.
A method for reconstructing an image, comprising:
- receiving a sinogram input;
- performing a first denoising on the sinogram input without reconstructing the sinogram input using a first trained artificial neural network;
- performing an initial reconstruction or multiple partial reconstructions of the sinogram input to generate one or more intermediate image;
- iteratively processing the one or more intermediate images or a transform of the one or more intermediate images, wherein the iterative processing comprises performing at least an iterative datafit operation comparing an output of an iterative loop with an input to the iterative loop;
- performing a second denoising operation on the one or more intermediate images in addition to the datafit operation, wherein the second denoising operation is performed using a second trained artificial neural network, and wherein the datafit operation and the second denoising operation are combined within the iterative loop;
- and
- upon completion of the iterative processing, reconstructing a final image and outputting the final image for review, storage, or further processing.
A method for reconstructing an image, comprising:
- receiving a sinogram input;
- performing a first denoising operation on the sinogram input without reconstructing the sinogram input using a first trained artificial neural network;
- reconstructing the sinogram input to generate one or more reconstructed images;
- performing a second denoising operation in the image domain using a second trained artificial neural network;
- iteratively processing the one or more reconstructed images or one or more intermediate sinograms generated from the one or more reconstructed images, wherein the iterative processing comprises performing at least an iterative datafit operation comparing an output of an iterative loop with an input to the iterative loop, wherein the datafit operation and the second denoising operation are combined within the iterative loop;
- and upon completion of the iterative processing, outputting a final image for review, storage, or further processing.
Dependencies are inferred from the claim text.