Systems and methods for segmentation of intra-patient medical images

1. A system for segmenting medical images, the system comprising:

a database configured to store a plurality of medical images acquired by an image acquisition device; and

a processor, configured to:

obtain one or more visual attributes of each image point of a plurality of image points selected from a first medical image of the plurality of medical images, the first medical image having a corresponding first structure label map;

for each image point of the plurality of image points, train a machine learning technique using the obtained one or more visual attributes to predict a structure label that matches the first structure label map corresponding to the respective image point, wherein the trained machine learning technique is applied to a second medical image to determine a second structure label map associated with the second medical image, wherein the machine learning technique comprises a convolutional neural network;

register the first medical image to the second medical image; and

train the convolutional neural network using the registered first medical image and the corresponding first structure label map.

2. The system according to claim 1 , wherein at least one of the one or more visual attributes includes an image intensity, an image texture, an image patch, or a curvature of an intensity profile.

3. The system according to claim 1 , wherein the first structure label map includes expert structure labels identified for the first medical image.

4. The system according to claim 1 , wherein the processor is further configured to determine the first structure label map for the first medical image using a population-trained classifier model.

5. The system according to claim 1 , wherein the processor is further configured to repeat the training for all image points in the first medical image.

6. The system according to claim 1 , wherein the machine learning technique further comprises a Random Forests model.

7. The system according to claim 1 , wherein the processor is further configured to identify at least one feature in the second medical image, and apply the machine learning technique to the at least one feature.

8. The system according to claim 7 , wherein the at least one feature is of a same type as the one or more visual attributes used to train the machine learning technique.

9. The system according to claim 1 , wherein the processor is further configured to:

obtain one or more visual attributes of a given image point from the second medical image;

apply the trained machine learning technique to the obtained one or more visual attributes of the given image point from the second medical image;

predict a structure label for the given image point based on applying the trained machine learning technique to the obtained one or more visual attributes of the given image point; and

provide a segmented image corresponding to the second medical image based on the predicted structure label, the structure label corresponding to the second structure label map.

10. The system according to claim 1 , wherein the processor is configured to determine the second structure label map prior to a radiotherapy treatment delivery.

11. A computer-implemented method for segmenting medical images, the method comprising the following operations performed by at least one processor:

obtaining one or more visual attributes of each image point of a plurality of image points selected from a first medical image of the plurality of medical images, the first medical image having a corresponding first structure label map;

for each image point of the plurality of image points, training a machine learning technique using the obtained one or more visual attributes to predict a structure label that matches the first structure label map corresponding to the respective image point, wherein the trained machine learning technique is applied to a second medical image to determine a second structure label map associated with the second medical image, wherein the machine learning technique comprises a convolutional neural network;

registering the first medical image to the second medical image; and

training the convolutional neural network using the registered first medical image and the corresponding first structure label map.

12. The method according to claim 11 , wherein at least one of the one or more visual attributes includes an image intensity, an image texture, an image patch, or a curvature of an intensity profile.

13. The method according to claim 11 , wherein the operations further comprise determining the first structure label map for the first medical image using a population-trained classifier model.

14. The method according to claim 11 , the first structure label map is registered to the second medical image using an atlas-based segmentation method.

15. The method according to claim 11 , wherein the operations further comprise training for all image points in the first medical image.

16. The method according to claim 11 , wherein the operations further comprise:

obtaining one or more visual attributes of a given image point from the second medical image;

applying the trained machine learning technique to the obtained one or more visual attributes of the given image point from the second medical image;

predicting a structure label for the given image point based on applying the trained machine learning technique to the obtained one or more visual attributes of the given image point; and

providing a segmented image corresponding to the second medical image based on the predicted structure label, the structure label corresponding to the second structure label map.

17. A non-transitory computer-readable medium containing instructions that, when executable by a processor, cause the processor to perform a method for segmenting medical images, the method comprising:

obtaining one or more visual attributes of each image point of a plurality of image points selected from a first medical image of the plurality of medical images, the first medical image having a corresponding first structure label map; and

for each image point of the plurality of image points, training a machine learning technique using the obtained one or more visual attributes to predict a structure label that matches the first structure label map corresponding to the respective image point, wherein the trained machine learning technique is applied to a second medical image to determine a second structure label map associated with the second medical image, wherein the machine learning technique comprises a convolutional neural network;

registering the first medical image to the second medical image; and

training the convolutional neural network using the registered first medical image and the corresponding first structure label map.