Deep learning based medical system and method for image acquisition

Patent metadata

Field	Value
Publication number	US-11763429-B2
Application number	US-202117325010-A
Country	US
Kind code	B2
Filing date	May 19, 2021
Priority date	May 19, 2021
Publication date	Sep 19, 2023
Grant date	Sep 19, 2023

Abstract

A medical imaging system having at least one medical imaging device providing image data of a subject is provided. The medical imaging system further includes a processing system programmed to train a deep learning (DL) network using a plurality of training images to predict noise in input data. The plurality of training images includes a plurality of excitation (NEX) images acquired for each line of k-space training data. The processing system is further programmed to use the trained DL network to determine noise in the image data of the subject and to generate a denoised medical image of the subject having reduced noise based on the determined noise in the image data.

First claim

The invention claimed is:

1. A medical imaging system comprising:
at least one medical imaging device providing image data of a subject;
a processing system programmed to:
train a deep learning (DL) network using a plurality of training images to predict noise in input data, wherein the plurality of training images includes a plurality of excitation images acquired for each line of k-space training data;
use the trained DL network to determine noise in the image data of the subject; and
generate a denoised medical image of the subject having reduced noise based on the determined noise in the image data;
wherein the plurality of excitation images includes at least one pair of excitation images;
wherein the processing system is programmed to train the DL network using input image data and target image data; and
wherein the input image data includes at least one of the excitation images of the at least one pair and the target image data includes at least one target noise signal derived from difference of two excitation images of the at least one pair.

2. The medical imaging system of claim 1 , wherein generating the denoised medical image comprises subtracting the noise of the image data from the image data of the subject.

3. The medical imaging system of claim 1 , wherein the DL network includes two cascaded Dense Blocks to predict noise in the input image.

4. The medical imaging system of claim 3 , wherein the target noise is a realization of the true noise present in the input image.

5. The medical imaging system of claim 3 , wherein an output image of the DL network is generated by adding the predicted noise in the input image.

6. The medical imaging system of claim 3 , wherein parameter values of dense blocks are adjusted to reduce differences between the output of the DL network and the target image of the DL network.

7. The medical imaging system of claim 6 , wherein a loss function is applied to the differences between the output of the DL network and the target image of the DL network to adjust the parameter values of Dense Blocks.

8. A method for imaging a subject comprising:
training a deep learning (DL) network using a plurality of training images to predict noise in input data, wherein the plurality of training images includes a plurality of excitation images acquired for each line of k-space training data using a magnetic resonance (MR) imaging device;
generating image data of the subject with the MR imaging device;
providing the image data of the subject as an input to the trained deep learning network model to determine noise in the image data of the subject;
generating a denoised medical image of the subject having reduced artifacts based on the determined noise in the image data;
wherein the plurality of excitation images includes at least one pair of excitation images;
wherein the DL network is trained using input image data and target image data; and
wherein the input image data includes at least one of the excitation images of the at least one pair and the target image data includes at least one target noise signal derived from difference of two excitation images of the at least one pair.

9. The method of claim 8 , wherein the DL network includes two cascaded Dense Blocks to predict noise in the input image.

10. The method of claim 9 , wherein the method further comprises adjusting parameter values of dense blocks to reduce differences between the output of the DL network and the target image of the DL network.

11. The method of claim 10 , wherein adjusting the parameter values of the Dense Blocks comprises applying a loss function to assess DL input and output mismatch.

12. The method of claim 8 , wherein generating the denoised medical image comprises subtracting the noise of the image data from the image data of the subject.

Assignees

• Ge Prec Healthcare Llc

Inventors

• Chatterjee Sudhanya
• Shanbhag Dattesh Dayanand

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• A61B5/055 Primary

  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• A61B5/0033

  Features or image-related aspects of imaging apparatus, e.g. for MRI, optical tomography or impedance tomography apparatus; Arrangements of imaging apparatus in a room · CPC title