Systems and methods for processing medical images with multi-layer perceptron neural networks

Patent metadata

Field	Value
Publication number	US-12412324-B2
Application number	US-202318095149-A
Country	US
Kind code	B2
Filing date	Jan 10, 2023
Priority date	Jan 10, 2023
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

Described herein are systems, methods, and instrumentalities associated with using a multi-layer perceptron (MLP) neural network to process medical images of an anatomical structure. The processing may include padding an input image in accordance with the training of the MLP neural network, splitting the input image (e.g., the padded input image) into patches of a same size, and processing the patches through the MLP neural network over one or more iterations. During an iteration of the processing, the patches may be processed separately and re-combined into an intermediate image before the intermediate image is shifted to concatenate portions of the image that are derived from different patches. This way, global features of the anatomical structure may be learned and used to improve the quality of the image generated by the MLP neural network, without incurring significant computation or memory costs.

First claim

What is claimed is:

1. An apparatus, comprising:
a processor configured to:
obtain an input image of an anatomical structure;
process the input image through a multi-layer perceptron (MLP) neural network over one or more iterations, wherein, during a first iteration of the one or more iterations, the processor is configured to:
divide the input image into at least a first patch having a specific image size and a second patch having the specific image size;
process the first patch and the second patch through the MLP neural network to derive a first intermediate image, wherein the first intermediate image includes a first portion derived based on the first patch and a second portion derived based on the second patch; and
shift the first intermediate image such that the first portion and the second portion of the first intermediate image are concatenated in the shifted first intermediate image; and

generate an output image of the anatomical structure in response to processing the input image of the anatomical structure through the MLP neural network, wherein the processor being configured to generate the output image comprises the processor being configured to reverse an effect caused by the shifting of the first intermediate image.

2. The apparatus of claim 1 , wherein, during a second iteration of the one or more iterations, the processor is configured to:
divide the shifted first intermediate image into at least a third patch having the specific image size and a fourth patch having the specific image size;
process the third patch and the fourth patch through the MLP neural network to derive a second intermediate image, wherein the second intermediate image includes a first portion derived based on the third patch and a second portion derived based on the fourth patch; and
shift the second intermediate image such that the first portion and the second portion of the second intermediate image are concatenated in the shifted second intermediate image.

3. The apparatus of claim 1 , wherein the processor being configured to reverse the effect caused by the shifting of the first intermediate image comprises the processor being configured to restore, in the output image of the anatomical structure, respective original positions of the first patch and the second patch as in the input image.

4. The apparatus of claim 1 , wherein the processor being configured to shift the first intermediate image comprises the processor being configured to append the first portion of the first intermediate image to the second portion of the first intermediate image.

5. The apparatus of claim 1 , wherein the processor is further configured to apply a padding to the input image of the anatomical structure so as to allow the first patch and the second patch to both have the specific image size.

6. The apparatus of claim 5 , wherein the padding increases a size of the input image to a multiple of the specific image size.

7. The apparatus of claim 5 , wherein the processor being configured to generate the output image of the anatomical structure comprises the processor being configured to remove the padding applied to the input image from the output image.

8. The apparatus of claim 1 , wherein the processor being configured to obtain the input image of the anatomical structure comprises the processor being configured to obtain under-sampled magnetic resonance (MR) data associated with the anatomical structure and generate the input image based on the under-sampled MR data.

9. The apparatus of claim 8 , wherein the processor being configured to generate the input image based on the under-sampled MR data comprises the processor being configured to generate the input image using a convolutional neural network trained for reconstructing the under-sampled MR data.

10. The apparatus of claim 1 , wherein the input image includes a multi-dimensional magnetic resonance (MR) image of the anatomical structure, the multi-dimensional MR image comprising a readout dimension and a phase-encoding dimension.

11. A method of processing medical images, the method comprising:
obtaining an input image of an anatomical structure;
processing the input image through a multi-layer perceptron (MLP) neural network over one or more iterations, wherein a first iteration of the one or more iterations comprises:
dividing the input image into at least a first patch having a specific image size and a second patch having the specific image size;
processing the first patch and the second patch through the MLP neural network to derive a first intermediate image, wherein the first intermediate image includes a first portion derived based on the first patch and a second portion derived based on the second patch; and shifting the first intermediate image such that the first portion and the second portion of the first intermediate image are concatenated in the shifted first intermediate image; and

generating an output image of the anatomical structure in response to processing the input image of the anatomical structure through the MLP neural network, wherein the generating of the output image comprises reversing an effect caused by the shifting of the first intermediate image.

12. The method of claim 11 , wherein a second iteration of the one or more iterations comprises:
dividing the shifted first intermediate image into at least a third patch having the specific image size and a fourth patch having the specific image size;
processing the third patch and the fourth patch through the MLP neural network to derive a second intermediate image, wherein the second intermediate image includes a first portion derived based on the third patch and a second portion derived based on the fourth patch; and
shifting the second intermediate image such that the first portion and the second portion of the second intermediate image are concatenated in the shifted second intermediate image.

13. The method of claim 11 , wherein reversing the effect caused by the shifting of the first intermediate image comprises restoring, in the output image of the anatomical structure, respective original positions of the first patch and the second patch as in the input image.

14. The method of claim 11 , wherein shifting the first intermediate image comprises appending the first portion of the first intermediate image to the second portion of the first intermediate image.

15. The method of claim 11 , further comprising applying a padding to the input image of the anatomical structure so as to allow the first patch and the second patch to both have the specific image size, wherein generating the output image of the anatomical structure comprises removing the padding applied to the input image from the output image.

16. The method of claim 11 , wherein obtaining the input image of the anatomical structure comprises obtaining under-sampled magnetic resonance (MR) data associated with the anatomical structure and generating the input image using a convolutional neural network trained for reconstructing the under-sampled MR data.

17. The method of claim 11 , wherein the input image includes a multi-dimensional magnetic resonance (MR) image of the anatomical structure, the multi-dimensional MR image comprising a readout dimension and a phase-encoding dimension.

18. A non-transitory computer-readable medium comprising instructions that, when executed by a processor included in a computing device, cause the processor to implement the method of claim 11 .

Assignees

• Shanghai United Imaging Intelligence Co Ltd

Inventors

• Chen Zhang
• Sun Shanhui
• Chen Xiao
• Liu Yikang
• Chen Terrence
• Zhang Chi

Classifications

• G06T12/30 Primary

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

• G06N3/045

  Combinations of networks · CPC title

• G06T5/50

  using two or more images, e.g. averaging or subtraction · CPC title

• G06T2211/424

  Iterative · CPC title

• G06T2207/30016

  Brain · CPC title