Image segmentation using neural network method

Patent metadata

Field	Value
Publication number	US-9947102-B2
Application number	US-201615248490-A
Country	US
Kind code	B2
Filing date	Aug 26, 2016
Priority date	Aug 26, 2016
Publication date	Apr 17, 2018
Grant date	Apr 17, 2018

Abstract

The present disclosure relates to systems, methods, devices, and non-transitory computer-readable storage medium for segmenting three-dimensional images. In one implementation, a computer-implemented method for segmenting a three-dimensional image is provided. The method may include receiving a three-dimensional image acquired by an imaging device, and selecting a plurality of stacks of adjacent two-dimensional images from the three-dimensional image. The method may further include segmenting, by a processor, each stack of adjacent two-dimensional images using a neural network model. The method may also include determining, by the processor, a label map for the three-dimensional image by aggregating the segmentation results from the plurality of stacks.

First claim

What is claimed is:

1. A computer-implemented method for segmenting a three-dimensional medical image, the method comprising:
receiving the three-dimensional medical image acquired by an imaging device;
selecting a plurality of stacks of adjacent two-dimensional images from the three-dimensional medical image;
segmenting, by a processor, each stack of adjacent two-dimensional images using a neural network model; and
determining, by the processor, a label map for the three-dimensional medical image by aggregating the segmentation results from the plurality of stacks.

2. The method of claim 1 , further including training the neural network model using at least one three-dimensional medical training image.

3. The method of claim 2 , wherein training the neural network model includes determining parameters of at least one convolution filter used in the neural network model.

4. The method of claim 1 , wherein each stack includes an odd number of two-dimensional images, and wherein segmenting the stack of adjacent two-dimensional images includes determining a label map for the two-dimensional image in the middle of the stack.

5. The method of claim 1 , wherein each stack includes an even number of two-dimensional images, and wherein segmenting the stack of adjacent two-dimensional images includes determining a label map for at least one of the two two-dimensional images in the middle of the stack.

6. The method of claim 1 , wherein the adjacent two-dimensional images are in the same plane and carry dependent structure information in an axis orthogonal to the plane.

7. The method of claim 1 , wherein the neural network model is a deep convolutional neural network model.

8. The method of claim 1 , wherein the three-dimensional medical image is a medical image indicative of anatomical structures of a patient, wherein the label map associates an anatomic structure to each voxel of the three-dimensional medical image.

9. A device for segmenting a three-dimensional medical image, the device comprising:
an input interface that receives the three-dimensional medical image acquired by an imaging device;
at least one storage device configured to store the three-dimensional medical image; and
an image processor configured to:
select a plurality of stacks of adjacent two-dimensional images from the three-dimensional medical image;
segment each stack of adjacent two-dimensional images using a neural network model; and
determine a label map for the three-dimensional medical image by aggregating the segmentation results from the plurality of stacks.

10. The device of claim 9 , wherein the image processor is further configured to train the neural network model using at least one three-dimensional medical training image.

11. The device of claim 10 , wherein the image processor is further configured to determine parameters of at least one convolution filter used in the neural network model.

12. The device of claim 9 , wherein each stack includes an odd number of two-dimensional images, and wherein the image processor is further configured to determine a label map for the two-dimensional image in the middle of the stack.

13. The device of claim 9 , wherein each stack includes an even number of two-dimensional images, and wherein the image processor is further configured to determine a label map for at least one of the two two-dimensional image in the middle of the stack.

14. The device of claim 9 , wherein the adjacent two-dimensional images are in the same plane and carry dependent structure information in an axis orthogonal to the plane.

15. The device of claim 9 , wherein the neural network model is a deep convolutional neural network model.

16. The device of claim 9 , wherein the three-dimensional medical image is a medical image indicative of anatomical structures of a patient, wherein the label map associates an anatomic structure to each voxel of the three-dimensional medical image.

17. A non-transitory computer-readable medium containing instructions that, when executable by at least one processor, cause the at least one processor to perform a method for segmenting a three-dimensional medical image, the method comprising:
receiving the three-dimensional medical image acquired by an imaging device;
selecting a plurality of stacks of adjacent two-dimensional images from the three-dimensional medical image;
segmenting each stack of adjacent two-dimensional images using a neural network model; and
determining a label map for the three-dimensional medical image by aggregating the segmentation results from the plurality of stacks.

18. The non-transitory computer-readable medium of claim 17 , wherein the method further includes training the neural network model using at least one three-dimensional medical training image.

19. The non-transitory computer-readable medium of claim 18 , wherein training the neural network model includes determining parameters of at least one convolution filter used in the neural network model.

20. The non-transitory computer-readable medium of claim 17 , wherein each stack includes an odd number of two-dimensional images, and wherein segmenting the stack of adjacent two-dimensional images includes determining a label map for the two-dimensional image in the middle of the stack.

21. The non-transitory computer-readable medium of claim 17 , wherein each stack includes an even number of two-dimensional images, and wherein segmenting the stack of adjacent two-dimensional images includes determining a label map for at least one of the two two-dimensional images in the middle of the stack.

22. The non-transitory computer-readable medium of claim 17 , wherein the adjacent two-dimensional images are in the same plane and carry dependent structure information in an axis orthogonal to the plane.

23. The non-transitory computer-readable medium of claim 17 , wherein the neural network model is a deep convolutional neural network model.

24. The non-transitory computer-readable medium of claim 17 , wherein the three-dimensional medical image is a medical image indicative of anatomical structures of a patient, wherein the label map associates an anatomic structure to each voxel of the three-dimensional medical image.

Assignees

• Elekta Inc

Inventors

• Xu Jiaofeng
• Han Xiao

Classifications

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• A61N5/10

  X-ray therapy; Gamma-ray therapy; Particle-irradiation therapy (A61N5/01 takes precedence) · CPC title

• G06F18/2411

  based on the proximity to a decision surface, e.g. support vector machines · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F18/00

  Pattern recognition · CPC title