Real-time patient motion monitoring using a magnetic resonance linear accelerator (MR-LINAC)
Systems and techniques may be used to estimate a real-time patient state during a radiotherapy treatment using a magnetic resonance linear accelerator (MR-Linac). For example, a method may include generating a dictionary of expanded potential patient measurements and corresponding potential patient states using a preliminary motion model. The method may include training, using a machine learning technique, a correspondence motion model relating an input patient measurement to an output patient state using the dictionary. The method may include estimating, using a processor, the patient state corresponding to a 2D MR image using the correspondence motion model. The method may include directing radiation therapy, using the MR-Linac, to a target according to the patient state.
1. A method for estimating a real-time patient state during a radiotherapy treatment using a magnetic resonance linear accelerator (MR-Linac), the method comprising:
generating a dictionary of simulated expanded potential patient measurements and corresponding simulated potential patient states using a preliminary motion model and based on an initial actual measurement and a corresponding initial actual patient state;
training, using a machine learning technique, a correspondence motion model relating an input patient measurement to an output patient state using the dictionary;
receiving a real-time stream of images from an image acquisition device;
estimating, using the processor, the patient state corresponding to an image of the real-time stream of images using the correspondence motion model; and
directing radiation therapy, using a treatment device coupled to the image acquisition device, to a target according to the patient state.
2. The method of claim 1 , wherein the simulated expanded potential patient measurements include deformations of a 3D patient image, and wherein the deformations include deformation vector fields (DVFs) calculated using a deformable registration algorithm.
3. The method of claim 1 , wherein the simulated expanded potential patient measurements are generated from a 4D image, the 4D image including a 4D CT, a 4D CBCT, a 4D MM, a 4D PET, or a 4D ultrasound image.
4. The method of claim 1 , wherein the correspondence motion model includes a deformation vector field (DVF) as a function of one or more parameters, the one or more parameters determined by reducing dimensionality of a preliminary DVF calculated between two or more phases of a 4D image and a reference phase.
5. The method of claim 1 , wherein the simulated expanded potential patient measurements include a 2D projection image and are generated by using at least one of:
extracting a 2D slice from a 3D image, ray-tracing through a 3D image to generate a 2D projection image, simulating x-ray interactions with a 3D image using a Monte Carlo technique, using a collapsed cone convolution technique, using a superposition and convolution technique, using a generative adversarial network, a convolutional neural network, or a recurrent neural network.
6. The method of claim 1 , wherein the correspondence motion model is generated using a random forest regression, a linear regression, a polynomial regression, a regression tree, a kernel density estimation, a support vector regression algorithm, a convolutional neural network, or a recurrent neural network.
7. The method of claim 1 , further comprising outputting the patient state as two or more MR-like 3D images showing tissue contrast.
8. The method of claim 1 , further comprising generating the simulated expanded potential patient measurements by calculating a 2D deformation vector field (DVF) on a 2D input image.
9. The method of claim 8 , wherein generating the simulated expanded potential patient measurements includes performing a principal component analysis (PCA) analysis of the 2D input image.
10. The method of claim 9 , wherein generating the simulated expanded potential patient measurements includes registering the 2D input image to a reference 2D image, and using a deformable image registration technique to calculate the 2D DVF, and using a convolutional neural network (CNN) to estimate a 2D optical flow between the 2D input image and a 2D reference image to calculate the 2D DVF.
11. The method of claim 1 , wherein the real-time stream of images include 2D MR images, low resolution 3D MR images, or 1D navigators.
12. A method for real-time tracking of a target, during a radiotherapy treatment using a magnetic resonance linear accelerator (MR-Linac), the method comprising:
generating a dictionary of simulated expanded potential patient measurements and corresponding simulated potential patient states using a preliminary motion model and based on an initial actual measurement and a corresponding initial actual patient state;
training, using a machine learning technique, a correspondence motion model relating an input patient measurement to an output patient state using the dictionary;
receiving a real-time stream of images from an image acquisition device;
estimating, using the processor, patient states corresponding to images in the real-time stream of images using the correspondence motion model;
tracking a radiation therapy target of a patient in real-time using the patient states; and
outputting tracking information for the radiation therapy target for display on a display device.
13. The method of claim 12 , wherein the simulated expanded potential patient measurements include deformations of a 3D patient image, and wherein the deformations include deformation vector fields (DVFs) calculated using a deformable registration algorithm.
14. The method of claim 12 , wherein the simulated expanded potential patient measurements are generated from a 4D image, the 4D image including a 4D CT, a 4D CBCT, a 4D MM, a 4D PET, or a 4D ultrasound image.
15. The method of claim 12 , wherein the correspondence motion model includes a deformation vector field (DVF) as a function of one or more parameters, the one or more parameters determined by reducing dimensionality of a preliminary DVF calculated between two or more phases of a 4D image and a reference phase.
16. The method of claim 12 , wherein the simulated expanded potential patient measurements include a 2D projection image and are generated by using at least one of:
extracting a 2D slice from a 3D image, ray-tracing through a 3D image to generate a 2D projection image, simulating x-ray interactions with a 3D image using a Monte Carlo technique, using a collapsed cone convolution technique, using a superposition and convolution technique, using a generative adversarial network, a convolutional neural network, or a recurrent neural network.
17. The method of claim 12 , wherein the correspondence motion model is generated using a random forest regression, a linear regression, a polynomial regression, a regression tree, a kernel density estimation, a support vector regression algorithm, a convolutional neural network, or a recurrent neural network.
18. The method of claim 12 , further comprising outputting the patient state as two or more MR-like 3D images showing tissue contrast.
19. The method of claim 12 , further comprising generating the simulated expanded potential patient measurements by calculating a 2D deformation vector field (DVF) on a 2D input image, and performing a principal component analysis (PCA) analysis of the 2D input image.
20. The method of claim 12 , wherein the real-time stream of images include 2D MR images, low resolution 3D MR images, or 1D navigators.