System and methods for image segmentation using convolutional neural network
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 the three-dimensional image acquired by an imaging device, and creating a first stack of two-dimensional images from a first plane of the three-dimensional image and a second stack of two-dimensional images from a second plane of the three-dimensional image. The method may further include segmenting, by a processor, the first stack and the second stack of two-dimensional images using at least one 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 first stack and second stack.
1. A computer-implemented method for training a machine learning model to segment a three-dimensional medical image to perform radiotherapy treatment planning, the method comprising:
receiving, by a processor, the three-dimensional medical image to be segmented;
selecting a first stack of two-dimensional images corresponding to a first plane of the three-dimensional medical image;
extracting one or more features of the first stack of two-dimensional images;
generating a training label map for an image in the first stack of two-dimensional images based on the extracted one or more features; and
training the machine learning model to generate a new label map for a subsequently received three-dimensional medical image based on a comparison of the generated training label map with a ground truth label map corresponding to the first stack of two-dimensional images, wherein a radiotherapy treatment plan is generated based on the trained machine learning model.
2. The method of claim 1 , further comprising:
selecting a second stack of two-dimensional images from a second plane of the three-dimensional medical image; and
generating a training label map for an image in the second stack of two-dimensional images.
3. The method of claim 2 , further comprising:
combining the first stack of two-dimensional image and the second stack of two-dimensional images; and
segmenting the combined stack of two-dimensional images using the machine learning model.
4. The method of claim 2 , wherein at least one of the two-dimensional images in the first stack are adjacent or the first and second stacks are adjacent.
5. The method of claim 1 further comprising computing an error map based on the comparison of the generated training label map with the ground truth label map.
6. The method of claim 5 further comprising updating a parameter of the machine learning model based on the error map.
7. The method of claim 1 , wherein the machine learning model comprises a deep convolutional neural network model.
8. The method of claim 1 further comprising determining whether a stopping criterion has been satisfied.
9. The method of claim 8 further comprising:
providing the machine learning model for application to the subsequently received three-dimensional medical image in response to determining that the stopping criterion has been satisfied; and
selecting another stack of two-dimensional images in response to determining that the stopping criterion has not been satisfied.
10. The method of claim 9 , wherein the first stack and the another stack have at least one overlapping image.
11. The method of claim 8 , wherein the stopping criterion includes at least one of a maximum number of iterations, a specified accuracy.
12. The method of claim 8 , wherein determining whether the stopping criterion has been satisfied comprises determining whether a loss function is smaller than a threshold.
13. A non-transitory computer readable medium comprising non-transitory computer readable instructions for training a machine learning model to segment a three-dimensional medical image to perform radiotherapy treatment planning, the non-transitory computer readable instructions comprising instructions for:
receiving, by a processor, the three-dimensional medical image to be segmented;
selecting a first stack of two-dimensional images corresponding to a first plane of the three-dimensional medical image;
extracting one or more features of the first stack of two-dimensional images;
generating a training label map for an image in the first stack of two-dimensional images based on the extracted one or more features; and
training the machine learning model to generate a new label map for a subsequently received three-dimensional medical image based on a comparison of the generated training label map with a ground truth label map corresponding to the first stack of two-dimensional images, wherein a radiotherapy treatment plan is generated based on the trained machine learning model.
14. The non-transitory computer readable medium of claim 13 , further comprising instructions for:
selecting a second stack of two-dimensional images from a second plane of the three-dimensional medical image; and
generating a training label map for an image in the second stack of two-dimensional images.
15. The non-transitory computer readable medium of claim 14 , further comprising instructions for:
combining the first stack of two-dimensional image and the second stack of two-dimensional images; and
segmenting the combined stack of two-dimensional images using the machine learning model.
16. The non-transitory computer readable medium of claim 14 , wherein at least one of the two-dimensional images in the first stack are adjacent or the first and second stacks are adjacent.
17. The non-transitory computer readable medium of claim 13 further comprising instructions for determining whether a stopping criterion has been satisfied.
18. The non-transitory computer readable medium of claim 17 further comprising instructions for:
providing the machine learning model for application to the subsequently received three-dimensional medical image in response to determining that the stopping criterion has been satisfied; and
selecting another stack of two-dimensional images in response to determining that the stopping criterion has not been satisfied.
19. A system to train a machine learning model to segment a three-dimensional medical image to perform radiotherapy treatment planning, the system comprising:
at least one storage device configured to store a three-dimensional medical image; and
a processor configured to:
select a first stack of two-dimensional images corresponding to a first plane of the three-dimensional medical image;
extract one or more features of the first stack of two-dimensional images;
generate a training label map for an image in the first stack of two-dimensional images based on the extracted one or more features; and
train the machine learning model to generate a new label map for a subsequently received three-dimensional medical image based on a comparison of the generated training label map with a ground truth label map corresponding to the first stack of two-dimensional images, wherein a radiotherapy treatment plan is generated based on the trained machine learning model.
20. The image-guided radio therapy system of claim 19 , wherein the machine learning model comprises a deep convolutional neural network model.