US 12,254,533 B2
Process for iteratively reconstructing images using deep learning
Mayank Patwari, Erlangen (DE); Ralf Gutjahr, Nuremberg (DE); and Rainer Raupach, Heroldsbach (DE)
Assigned to SIEMENS HEALTHINEERS AG, Forchheim (DE)
Filed by Siemens Healthcare GmbH, Erlangen (DE)
Filed on Feb. 11, 2022, as Appl. No. 17/669,578.
Claims priority of application No. 21159672 (EP), filed on Feb. 26, 2021.
Prior Publication US 2022/0277497 A1, Sep. 1, 2022
Int. Cl. G06N 3/08 (2023.01); G06N 3/045 (2023.01); G06T 11/00 (2006.01); A61B 6/03 (2006.01)
CPC G06T 11/003 (2013.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); A61B 6/032 (2013.01); G06T 2207/10081 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2211/421 (2013.01); G06T 2211/424 (2013.01)] 22 Claims
OG exemplary drawing
 
8. A data processing system comprising:
an input module configured to receive at least one of projections or volumes of an imaging procedure;
an output module configured to be communicatively connected with, and to provide, volumes to external modules;
a projection bilateral filter configured to filter the projections;
a volume filter configured to filter the volumes;
a backward-projector communicatively connected to the volume filter, and configured to backward-project volumes from the projections;
a forward-projector communicatively connected to the volume filter, and configured to forward-project projections from the volumes;
a first trained neural network communicatively connected to the forward-projector, and configured to receive data of projections as input, and to provide at least one of a spatial projection filter parameter or an intensity projection filter parameter of the projection bilateral filter as output;
a second trained neural network communicatively connected to the backward-projector, and configured to receive data of a volume as input, and to provide at least one of a spatial volume filter parameter or an intensity volume filter parameter of the volume filter as output;
wherein the input module is communicatively connected to at least one of the first trained neural network, the second trained neural network, the backward-projector, or the forward-projector;
wherein the first trained neural network is configured to automatically tune at least one of the spatial projection filter parameter or the intensity projection filter parameter based on the data of the projections or on data of the forward-projected projections;
wherein the projection bilateral filter is configured to filter the projections using at least one projection filter parameter;
wherein the backward-projector is configured to backward-project a volume from the projections or from the filtered projections, using a filtered back-projection;
wherein the second trained neural network is configured to automatically tune at least one of the spatial volume filter parameter or the intensity volume filter parameter of the volume filter based on the data of the volume or on the backward-projected volume;
wherein the volume filter is configured to filter the volume using at least one volume filter parameter;
wherein the forward-projector is configured to forward-project projections from the volume or from the filtered volume; and
wherein the output module is communicatively connected to the volume filter, and configured to provide the filtered volume to the external modules, in case the filtered volume meets a quality criterion.