Real-time motion monitoring using deep learning
Systems and methods may be used for estimating instantaneous patient motion (a patient state). The patient state may be estimated based on a 3D reference volume and a stream of images, for example from an image acquisition device. The stream of images may be received in real-time, for example during a radiation therapy treatment. An example method may include encoding the 3D reference volume using a 3D encoder branch of a patient state generator network, encoding the stream of images using a 2D encoder branch of the patient state generator network, and combining the encoded 3D reference volume and the encoded real-time stream of images. The method may include estimating a 3D spatial transform that maps the 3D reference volume to a current patient state by decoding the combined encoding using a 3D decoder branch of the patient state generator network.
1. A method for estimating a patient state comprising:
receiving a 3D reference volume;
receiving a real-time stream of images from an image acquisition device;
using a hardware processor to implement a patient state generator network to:
encode the 3D reference volume using a 3D encoder branch of the patient state generator network;
encode the real-time stream of images using a 2D encoder branch of the patient state generator network;
combine the encoded 3D reference volume and the encoded real-time stream of images; and
estimate a 3D spatial transform that maps the 3D reference volume to a current patient state using a 3D decoder branch of the patient state generator network.
2. The method of claim 1 , wherein the 3D spatial transform is a deformation vector field (DVF).
3. The method of claim 1 , wherein the real-time stream of images is a stream of slices of magnetic resonance (MR) images or a kV projection image (X-ray).
4. The method of claim 1 , wherein the 3D reference volume is a 3D magnetic resonance (MR) image or a 3D computed tomography (CT) image.
5. The method of claim 1 , wherein the patient state generator network is trained as a single neural network.
6. The method of claim 5 , wherein the 3D reference volume and the real-time stream of images are used as input to the single neural network, and wherein the 3D spatial transform is an output of the single neural network.
7. The method of claim 1 , wherein the 3D reference volume is captured prior to the real-time stream of images.
8. The method of claim 1 , wherein the real-time stream of images is received during a radiation therapy treatment.
9. A method for estimating a patient state comprising:
using a hardware processor to implement a patient state generator network to:
encode a 3D reference volume using a 3D encoder branch of the patient state generator network;
encode a stream of images received during a radiation therapy treatment using a 2D encoder branch of the patient state generator network;
combine the encoded 3D reference volume and the encoded real-time stream of images; and
estimate a 3D spatial transform that maps the 3D reference volume to a current patient state using a 3D decoder branch of the patient state generator network.
10. The method of claim 9 , wherein the 3D spatial transform is a deformation vector field (DVF).
11. The method of claim 9 , wherein the stream of images is a stream of slices of magnetic resonance (MR) images or a kV projection image (X-ray).
12. The method of claim 9 , wherein the 3D reference volume is a 3D magnetic resonance (MR) image or a 3D computed tomography (CT) image.
13. The method of claim 9 , wherein the patient state generator network is trained as a single neural network.
14. The method of claim 13 , wherein the 3D reference volume and the stream of images are used as input to the single neural network, and wherein the 3D spatial transform is an output of the single neural network.
15. The method of claim 9 , wherein the 3D reference volume is captured prior to the radiation therapy treatment.
16. A system for estimating a patient state comprising:
a hardware processor;
memory including instructions, which when executed by the hardware processor, cause the hardware processor to implement a patient state generator network to:
encode a 3D reference volume using a 3D encoder branch of the patient state generator network;
encode the real-time stream of images using a 2D encoder branch of the patient state generator network;
combine the encoded 3D reference volume and the encoded real-time stream of images; and
estimate a 3D spatial transform that maps the 3D reference volume to a current patient state using a 3D decoder branch of the patient state generator network.
17. The method of claim 16 , wherein the 3D spatial transform is a deformation vector field (DVF).
18. The method of claim 16 , wherein the real-time stream of images is a stream of slices of magnetic resonance (MR) images or a kV projection image (X-ray).
19. The method of claim 16 , wherein the 3D reference volume is a 3D magnetic resonance (MR) image or a 3D computed tomography (CT) image.
20. The method of claim 16 , wherein the patient state generator network is trained as a single neural network.