Method, system and device of image segmentation

1 . A method of image segmentation, comprising:

acquiring image information, the image information comprising an image to be segmented, a template, and a segmentation mask of the template; and

obtaining a segmentation result of the image to be segmented based on the image information by an image segmentation model, and the image segmentation model comprising an image transforming network and a segmenting network;

wherein acquiring the image information comprises:

acquiring a plurality of moving images and corresponding partition masks;

registering the plurality of moving images with a standard space, and obtaining a plurality of registered images corresponding to the plurality of moving images; and

generating the template based on the plurality of the registered images.

2 . The method of claim 1 , wherein the obtaining the segmentation result of the image to be segmented based on the image information by the image segmentation model comprises:

obtaining a transformed image based on the image to be segmented and the template by the image transforming network; and

obtaining the segmentation result based on the transformed image and the segmentation mask of the template by the segmenting network.

3 . The method of claim 1 , wherein the image to be segmented comprises symmetry information, and the image segmentation model is configured to obtain the segmentation result based on the symmetry information.

4 . The method of claim 1 , wherein the acquiring the image information further comprising:

mapping the partition masks corresponding to the plurality of moving images to the standard space based on a registration relationship of the plurality of moving images and the standard space, and obtaining a mapped result; and

obtaining the segmentation mask of the template based on the mapped result.

5 . The method of claim 1 , wherein prior to the obtaining the segmentation result of the image to be segmented based on the image information by the image segmentation model, the method further comprises training the image segmentation model, and the training the image segmentation model comprises:

acquiring sample image information, the sample image information comprising a sample image, a sample template, and a sample segmentation mask; and

training the image segmentation model based on the sample image information.

6 . The method of claim 5 , wherein the training the image segmentation model based on the sample image information comprises:

inputting a sample image and a sample template to the image transforming network to obtain an affine transformation field and a rigid transformation field;

obtaining an affinely transformed image based on the affine transformation field and the sample image;

obtaining a transformed image based on the rigid transformation field and the sample image;

obtaining a first loss function based on the transformed image;

obtaining a second loss function based on the sample template and the affinely transformed image; and

updating parameters of the image transforming network based on the first loss function and the second loss function.

7 . The method of claim 6 , wherein the training the image segmentation model based on the sample image information further comprises:

inputting the transformed image to the segmenting network to obtain transformation fields of multiple resolutions;

fusing the transformation fields of multiple resolutions to obtain a fused transformation field;

obtaining a transformed template and a transformed mask based on the sample template, the sample segmentation mask and the fused transformation field;

obtaining a third loss function based on the transformed image and the transformed template;

obtaining a fourth loss function based on the transformed mask and a mirror flip of the transformed mask; and

updating parameters of the segmenting network based on the third loss function and the fourth loss function.

8 . The method of claim 6 , wherein the training the image segmentation model based on the sample image information further comprises:

inputting the transformed image and the sample template to the segmenting network to obtain transformation fields of multiple resolutions;

fusing the transformation fields of multiple resolutions to obtain a fused transformation field;

obtaining a transformed template and a transformed mask based on the sample template, the sample segmentation mask and the fused transformation field;

obtaining a third loss function based on the transformed image and the transformed template;

obtaining a fourth loss function based on the transformed mask and a mirror flip of the transformed mask; and

updating parameters of the segmenting network based on the third loss function and the fourth loss function.

9 . The method of claim 7 , wherein after the training the image segmentation model, the method further comprises testing the image segmentation model, and the testing the image segmentation model comprises:

inputting the image to be segmented together with the template to the image transforming network to obtain rotation and translation quantities, and performing a rigid transformation to obtain a transformed image;

inputting the transformed image separately to the segmenting network to obtain a registered transformation field;

processing the segmentation mask of the template by using the registered transformation field to obtain the segmentation result of the transformed image; and

taking the segmentation result as an input of an automatic scoring.

10 . The method of claim 8 , wherein after the training the image segmentation model, the method further comprises testing the image segmentation model, and the testing the image segmentation model comprises:

inputting the image to be segmented together with the template to the image transforming network to obtain rotation and translation quantities, and performing a rigid transformation to obtain a transformed image;

inputting the transformed image and the template to the segmenting network to obtain a registered transformation field;

processing the segmentation mask of the template by using the registered transformation field to obtain the segmentation result of the transformed image; and

taking the segmentation result as an input of an automatic scoring.

11 . The method of claim 1 , wherein the obtaining the segmentation result of the image to be segmented based on the image information by the image segmentation model comprises:

obtaining a transformed image based on the image to be segmented and the template by the image transforming network; and

obtaining the segmentation result based on the transformed image, the template, and the segmentation mask of the template by the segmenting network.

12 . A system of image segmentation, comprising:

an image information acquiring module, configured to acquire image information, the image information comprising an image to be segmented, a template, and a segmentation mask of the template; and

an image segmentation module, configured to obtain a segmentation result of the image to be segmented based on the image information by an image segmentation model, and the image segmentation model comprising an image transforming network and a segmenting network; and

a template and mask acquisition module, configured to: acquire a plurality of moving images and corresponding partition masks; register the plurality of moving images with a standard space, and obtain a plurality of registered images corresponding to the plurality of moving images; and

generate the template based on the plurality of the registered images.

13 . The system of image segmentation of claim 12 , wherein the template and mask acquisition module is further configured to: map the partition masks corresponding to the plurality of moving images to the standard space based on a registration relationship of the plurality of moving images and the standard space, and obtain a mapped result; and obtain a segmentation mask based on the mapped result.

14 . The system of image segmentation of claim 12 , further comprising:

a training module, configured to: acquire sample image information, the sample image information comprising a sample image, a sample template, and a sample segmentation mask; and

train the image segmentation model based on the sample image information.

15 . The system of image segmentation of claim 14 , wherein the training module training the image segmentation model comprises the training module training an image transforming network and the training module training a segmenting network; and

the training module training the image transforming network comprises:

inputting a sample image and a sample template to the image transforming network to obtain an affine transformation field and a rigid transformation field;

obtaining an affinely transformed image based on the affine transformation field and the sample image;

obtaining a transformed image based on the rigid transformation field and the sample image;

obtaining a first loss function based on the transformed image;

obtaining a second loss function based on the sample template and the affinely transformed image; and

updating parameters of the image transforming network based on the first loss function and the second loss function.

16 . The system of image segmentation of claim 15 , wherein the training module training the segmenting network comprises:

inputting the transformed image to the segmenting network to obtain transformation fields of multiple resolutions;

fusing the transformation fields of multiple resolutions to obtain a fused transformation field;

obtaining a transformed template and a transformed mask based on the sample template, the sample segmentation mask and the fused transformation field;

obtaining a third loss function based on the transformed image and the transformed template;

obtaining a fourth loss function based on the transformed mask and a mirror flip of the transformed mask; and

updating parameters of the segmenting network based on the third loss function and the fourth loss function.

17 . A device of image segmentation, comprising a processor and a memory, wherein a computer program is stored in the memory, and when executing the computer program, the processor performs the method of claim 1 .

18 . A non-transitory computer-readable storage medium, wherein, computer instructions are stored in the storage medium, and when reading the computer instructions in the storage medium, the computer performs the method of claim 1 .