Calibration methods and systems for medical imaging

Patent metadata

Field	Value
Publication number	US-12386002-B2
Application number	US-202217664213-A
Country	US
Kind code	B2
Filing date	May 19, 2022
Priority date	Dec 14, 2021
Publication date	Aug 12, 2025
Grant date	Aug 12, 2025

Abstract

The present disclosure discloses a calibration method and system for medical imaging. The calibration method comprising: obtaining imaging data; dividing the imaging data into a plurality of patches; and calibrating the imaging data based on one or more target patches of the plurality of patches, wherein the one or more target patches is a part of the plurality of patches.

First claim

What is claimed is:

1. A calibration method implemented on a computing device having at least one processor and at least one storage device, comprising:
obtaining imaging data, the imaging data including K-space data collected by magnetic resonance imaging (MRI) device;
dividing the imaging data into a plurality of patches, each of the plurality of patches including multiple data points of the imaging data;
determining one or more target patches from the plurality of patches by processing the plurality of patches using at least one trained machine learning model, wherein the one or more target patches are part of the plurality of patches and each target patch includes one or more abnormal points;
calibrating the imaging data by calibrating the one or more target patches to obtain calibrated imaging data; and
generating a magnetic resonance image based on the calibrated imaging data.

2. The calibration method of claim 1 , wherein
the determining the one or more target patches from the plurality of patches by processing the plurality of patches using at least one trained machine learning model comprises;
processing the plurality of the patches using at least one trained first classification model to obtain an output of the at least one trained first classification model; and
determining the one or more target patches from the plurality of patches based on the output.

3. The calibration method of claim 2 , wherein the at least one trained first classification model includes a Vision Transformer deep learning model.

4. The calibration method of claim 2 , wherein the processing the plurality of patches using at least one trained first classification model to obtain an output of the at least one trained first classification model comprises:
for each patch of the plurality of patches,
determining a preprocessing parameter of the patch based on position information indicating a position of the patch in the imaging data; and
preprocessing the patch based on the preprocessing parameter of the patch to obtain a preprocessed patch; and

processing the plurality of preprocessed patches using the at least one trained first classification model to obtain an output of the at least one trained first classification model.

5. The calibration method of claim 2 , wherein the at least one trained first classification model includes a plurality of trained first classification models corresponding to different positions of K-space, the processing the plurality of patches using at least one trained first classification model comprises:
for each of the plurality of patches,
selecting, from the plurality of trained first classification models, a trained first classification model corresponding to the patch based on position information of the patch in the imaging data; and
processing the patch using the selected trained first classification model.

6. The calibration method of claim 2 , wherein the processing the plurality of the patches using at least one trained first classification model comprises:
processing the plurality of patches and position information of the plurality of patches in the imaging data using the at least one trained first classification model.

7. The calibration method of claim 6 , wherein
the at least one trained first classification model includes a convolutional neural network (CNN) model that includes a position encoding component integrated into a fully connected layer of the CNN model, the position information is processed using the position encoding component, or
the at least one trained first classification model includes a vision transformer (VIT) deep learning model that includes a transformer encoder layer, the position information is processed using the transformer encoder layer.

8. The calibration method of claim 2 , wherein the output of the at least one trained first classification model includes a probability value that each patch includes abnormal points,
the one or more target patches are determined from the plurality of patches based on the probability value and a preset classification threshold corresponding to each patch, preset classification thresholds corresponding to the patches at different positions in the imaging data are different.

9. The calibration method of claim 8 , wherein a preset classification threshold corresponding to a patch located in a central area of the K-space is greater than a preset classification threshold corresponding to a patch located in other areas of the K-space.

10. The calibration method of claim 1 , wherein the determining the one or more target patches from the plurality of patches by processing the plurality of patches using at least one trained machine learning model comprises:
generating a reconstructed image based on the K-space data;
determining whether the reconstructed image includes artifacts; and
determining the one or more target patches by processing the plurality of patches using at least one trained first classification model based on a determination result of whether the reconstructed image includes artifacts.

11. The calibration method of claim 10 , wherein the determining whether the reconstructed image includes artifacts comprises:
cropping at least one image block from the reconstructed image; and
determining whether the reconstructed image includes artifacts by processing the at least one image block using a trained second classification model.

12. The calibration method of claim 1 , wherein the calibrating the imaging data by calibrating the one or more target patches to obtain calibrated imaging data comprises:
obtaining one or more calibrated target patches by calibrating the one or more target patches using a trained artifact calibration model;
calibrating the imaging data based on the one or more calibrated target patches to obtain the calibrated imaging data.

13. The calibration method of claim 1 , wherein the calibrating the imaging data by calibrating the one or more target patches to obtain calibrated imaging data comprises:
segmenting the one or more abnormal points from the one or more target patches by using an abnormal point segmentation model; and
calibrating the imaging data based on a segmentation result of the one or more abnormal points to obtain the calibrated imaging data.

14. The calibration method of claim 1 , wherein the determining the one or more target patches from the plurality of patches by processing the plurality of patches using the at least one trained machine learning model comprises:
determining multiple sets of input data based on position information indicating positions of the plurality of patches in the imaging data; and
determining the one or more target patches by inputting the multiple sets of input data into at least one trained first classification model, respectively.

15. The calibration method of claim 1 , wherein the at least one trained machine learning model includes a trained patch calibration model, the one or more target patches are determined and calibrated by inputting the plurality of patches into the trained patch calibration model.

16. The calibration method of claim 15 , further comprising:
determining multiple sets of input data based on position information indicating positions of the plurality of patches in the imaging data, wherein the one or more target patches are determined and calibrated by inputting the multiple sets into the trained patch calibration model, respectively.

17. The calibration method of claim 1 , wherein the dividing the imaging data into a plurality of patches comprises:
obtaini

Assignees

• Shanghai United Imaging Healthcare Co Ltd

Inventors

• Tong Li
• Sun Tao
• Xin Yang

Classifications

• G06T12/30 Primary

  Image post-processing, e.g. metal artefact correction · CPC title

• G06T5/00

  Image enhancement or restoration · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• A61B6/582 Primary

  Calibration · CPC title

• A61B6/5258

  involving detection or reduction of artifacts or noise · CPC title