Reducing noise in CT images using synthetic data
AssigneeGE Precision Healthcare LLC
InventorsBrian E. Nett, Meghan L. Yue, Nicholas Mathew, Jonathan S. Maltz
Life of this patent
Estimating the term…
Abstract
The current disclosure provides methods and systems to reduce an amount of noise in image data. In one example, a method for creating synthetic computed tomography (CT) images for training a model to reduce an amount of noise in acquired CT images is proposed, comprising performing a tissue segmentation of reference images of an anatomical region of a subject to determine a set of different tissue types of the reference images; and generating synthetic CT images of the reference images by assigning CT image values of the synthetic CT images based on the different tissue types.
Claims14
An image processing system comprising:
- a processor communicably coupled to a non-transitory memory storing a neural network, the memory including instructions that when executed cause the processor to:
- receive a plurality of synthetic computed tomography (CT) images, each synthetic CT image generated by performing a tissue segmentation of a reference image of a plurality of reference images of a subject and assigning CT image values to types of tissues identified by the tissue segmentation;
- for each synthetic CT image of the plurality of synthetic CT images, add noise to the synthetic CT image to create a noisy version of the synthetic CT image;
- create a respective plurality of image pairs, each image pair including a synthetic CT image as a target, ground truth image, and a corresponding noisy version of the synthetic CT image as an input image;
- train the neural network using the image pairs;
- deploy the trained neural network to generate noise-reduced images from CT images acquired from a scanned subject;
- and
- display the noise-reduced images on a display device of the image processing system;
- wherein the plurality of reference images include a composite image generated from a combination of a plurality of magnetic resonance (MR) images, each MR image of the plurality of MR images acquired from a same slice of a scan of a patient using a different pulse sequence.
A method for an image processing system, comprising:
- acquiring a reference image of an anatomical region of a subject;
- segmenting tissues of the reference image;
- assigning voxel intensity values to different tissue types of the reference image to generate a first synthetic computed tomography (CT) image corresponding to the reference image;
- adding noise to the first synthetic CT image to generate a second, noisy synthetic CT image;
- creating a training image pair including the second synthetic CT image as an input image, and the first synthetic CT image as a ground truth image;
- training a convolutional neural network (CNN) to remove noise from synthetic CT images, using training data including the training image pair;
- using the trained CNN to remove noise from a real CT image acquired from a scanned subject;
- and
- displaying a noise-reduced version of the real CT image on a display device of the image processing system and/or storing the noise-reduced version in a memory of the image processing system;
- wherein the reference image is a composite image generated from a combination of a plurality of magnetic resonance (MR) images, each MR image of the plurality of MR images acquired from a same slice of a scan of a patient using a different pulse sequence.
Dependencies are inferred from the claim text.