US 12,462,347 B2
Generation of image denoising training data via photon-count splitting
Sen Wang, Stanford, CA (US); Yirong Yang, Stanford, CA (US); Zhye Yin, Niskayuna, NY (US); and Adam S. Wang, Palo Alto, CA (US)
Assigned to GE Precision Healthcare LLC, Waukesha, WI (US); and The Board of Trustees of the Leland Stanford Junior University, Stanford, CA (US)
Filed by GE Precision Healthcare LLC, Wauwatosa, WI (US); and The Board of Trustees of the Leland Stanford Junior University, Stanford, CA (US)
Filed on Oct. 27, 2022, as Appl. No. 18/050,111.
Prior Publication US 2024/0144442 A1, May 2, 2024
Int. Cl. G06T 5/00 (2024.01); A61B 6/00 (2006.01); A61B 6/42 (2024.01); G06T 5/70 (2024.01)
CPC G06T 5/70 (2024.01) [A61B 6/4241 (2013.01); A61B 6/5258 (2013.01); G06T 2207/10081 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system, comprising:
a memory configured to store computer-executable components; and
a processor that executes at least one of the computer-executable components that:
trains a deep learning neural network to perform denoising on computed tomography images, wherein the training comprises:
accessing a set of sinograms generated by at least one photon-counting computed tomography scanner;
splitting, via photon-wise binomial selection, the set of sinograms into a first reduced-photon-count set of sinograms and a second reduced-photon-count set of sinograms, wherein the photon-wise binomial selection probabilistically assigns photons recorded in the set of sinograms to the first reduced-photon-count set according to a defined probability value, and wherein the photon-wise binomial selection probabilistically assigns photons recorded in the set of sinograms to the second reduced-photon-count set according to a complement of the defined probability value;
converting, via image reconstruction, the first reduced-photon-count set of sinograms into at least one training input image, and the second reduced-photon-count set of sinograms into at least one training output image, wherein the photon-wise binomial selection causes the at least one training input image and the at least one training output image to be noise independent; and
training, using the at least one training input image and the at least one training output image, the deep learning neural network to perform the denoising.