Systems and methods for low-dose AI-based imaging
AssigneeMedtronic Navigation, Inc.
InventorsAndre D. A. Souza, Michael Philip Marrama, Patrick A. Helm, Mehdi Rahman, Kyo C. Jin, Michael D. Ketcha
Life of this patent
Estimating the term…
Abstract
A low-dose imaging method includes receiving a sparse image set of a portion of a patient's anatomy; up-sampling the sparse image set, in the sinogram domain and using a first neural network, to yield an up-sampled sinogram; generating, from the up-sampled sinogram, an initial reconstruction; and removing, from the initial reconstruction and using a second neural network, one or more artifacts in the initial reconstruction to yield a final output volume.
Claims20
A low-dose imaging method, comprising:
- receiving a sparse image set of a portion of a patient's anatomy, wherein the sparse image set comprises a plurality of globally-distributed artifacts;
- up-sampling the sparse image set, in a sinogram domain and using a first neural network trained in the sinogram domain, to yield an up-sampled sinogram in which the first neural network has filled in at least some of the plurality of globally-distributed artifacts;
- generating, from the up-sampled sinogram, an initial reconstruction;
- and
- removing, from the initial reconstruction and using a second neural network trained in a slice domain, one or more artifacts in the initial reconstruction to yield a final output volume.
A system for low-dose imaging, comprising:
- a cone-based computed tomography imaging device capable of selectively acquiring a normal dose image set and a low dose image set, the low dose image set comprising twenty-five percent or fewer images than the normal dose image set;
- at least one processor;
- and
- a memory storing data for execution by the at least one processor that, when executed, cause the at least one processor to:
- receive, from the imaging device, a low dose image set of a portion of a patient's anatomy comprising a surgical implant, wherein the low dose image set comprises a plurality of globally-distributed artifacts;
- up-sample the low dose image set, in a sinogram domain and using a first neural network trained in the sinogram domain, to yield an up-sampled sinogram in which the first neural network has filled in at least some of the plurality of globally-distributed artifacts;
- generate, from the up-sampled sinogram and using filtered back projection, an initial reconstruction;
- and
- remove, from the initial reconstruction and using a second neural network trained in a slice domain, one or more artifacts in the initial reconstruction to yield a final output volume.
Dependencies are inferred from the claim text.