Radiotherapy treatment plans using differentiable dose functions

1. A method for generating a radiotherapy treatment plan parameter, the method comprising:

receiving, by processor circuitry, radiotherapy treatment plan information;

processing, by the processor circuitry, the radiotherapy treatment plan information to estimate one or more radiotherapy treatment plan parameters based on a process that depends on an output of a subprocess that estimates a derivative of a dose calculation, wherein the derivative of the dose calculation is used in a machine learning model that is based on a loss function, wherein the derivative of the dose calculation is computed with respect to at least one of one or more radiation parameters or one or more geometry parameters of a radiotherapy treatment device, wherein processing the radiotherapy treatment plan information comprises processing the radiotherapy treatment plan information with the machine learning model to generate the one or more estimated radiotherapy treatment plan parameters, and wherein the machine learning model is trained, based on a plurality of training dose calculations, to establish a relationship between a plurality of training radiotherapy treatment plan information and a plurality of training radiotherapy treatment plan parameters; and

generating, by the processor circuitry, a radiotherapy treatment plan using the estimated one or more radiotherapy treatment plan parameters.

2. The method of claim 1 , wherein the radiotherapy treatment plan information includes at least one of a magnetic resonance (MR) image, a cone-beam computed tomography (CBCT) image, a computed tomography (CT) image, a dose distribution, a segmentation map or a distance map.

3. The method of claim 2 , wherein the estimated one or more radiotherapy treatment plan parameters comprises at least one of a synthetic computed tomography (sCT) image or a dose distribution.

4. The method of claim 1 , wherein the machine learning model includes a deep neural network, that is trained based on the plurality of training dose calculations using the plurality of training radiotherapy treatment plan information, wherein the plurality of training radiotherapy treatment plan information comprises at least one of a training magnetic resonance (MR) image, a training cone-beam computed tomography (CBCT) image, a training computed tomography (CT) image, a first training dose distribution, a training segmentation map or a training distance map, and wherein the plurality of training radiotherapy treatment plan parameters comprises at least one of a training synthetic computed tomography (sCT) image or a second training dose distribution.

5. The method of claim 1 , further comprising training the machine learning model by:

obtaining a first batch of training data pairs comprising a given set of training radiotherapy treatment plan information;

processing the given set of training radiotherapy treatment plan information with the machine learning model to generate an intermediate radiotherapy treatment plan parameter;

computing a derivative of the loss function based on the intermediate radiotherapy treatment plan parameter; and

updating parameters of the machine learning model based on the computed derivative of the loss function.

6. The method of claim 5 , wherein the first batch of training data pairs comprises a set of corresponding training radiotherapy treatment plan parameters associated with the given set of training radiotherapy treatment plan information, further comprising:

computing a first dose based on the set of corresponding training radiotherapy treatment plan parameters;

computing a second dose based on the intermediate radiotherapy treatment plan parameter; and

applying the first and second doses to the loss function before computing the derivative of the loss function.

7. The method of claim 1 , wherein the dose calculation includes at least one or a combination of a Monte Carlo simulation or a deterministic calculation using a point kernel convolution algorithm, a pencil kernel algorithm, or a Boltzmann equation solver.

8. The method of claim 1 , wherein the derivative of the dose calculation is a first-order derivative.

9. The method of claim 8 , wherein the first-order derivative is not a constant value.

10. The method of claim 1 , wherein the subprocess that estimates a derivative comprises an automatic differentiation process.

11. A non-transitory computer-readable medium comprising non-transitory computer-readable instructions, the computer-readable instructions comprising instructions for performing operations comprising:

receiving radiotherapy treatment plan information;

processing the radiotherapy treatment plan information to estimate one or more radiotherapy treatment plan parameters based on a process that depends on an output of a subprocess that estimates a derivative of a dose calculation, wherein the derivative of the dose calculation is used in a machine learning model that is based on a loss function, and wherein the derivative of the dose calculation is computed with respect to at least one of one or more radiation parameters or one or more geometry parameters of a radiotherapy treatment device, wherein processing the radiotherapy treatment plan information comprises processing the radiotherapy treatment plan information with the machine learning model to generate the one or more estimated radiotherapy treatment plan parameters, the machine learning model being trained, based on a plurality of training dose calculations, to establish a relationship between a plurality of training radiotherapy treatment plan information and a plurality of training radiotherapy treatment plan parameters; and

generating a radiotherapy treatment plan using the estimated one or more radiotherapy treatment plan parameters.

12. The non-transitory computer-readable medium of claim 11 , wherein the radiotherapy treatment plan information includes at least one of a magnetic resonance (MR) image, a cone-beam computed tomography (CBCT) image, a computed tomography (CT) image, a dose distribution, a segmentation map or a distance map.

13. The non-transitory computer-readable medium of claim 11 , wherein the machine learning model comprises a deep neural network that is trained based on the plurality of training dose calculations using the plurality of training radiotherapy treatment plan information, wherein the plurality of training radiotherapy treatment plan information comprises at least one of a training magnetic resonance (MR) image, a training cone-beam computed tomography (CBCT) image, a training computed tomography (CT) image, a first training dose distribution, a training segmentation map or a training distance map, and wherein the plurality of training radiotherapy treatment plan parameters comprises at least one of a training synthetic computed tomography (sCT) image or a second training dose distribution.

14. The non-transitory computer-readable medium of claim 11 , the operations further comprising training the machine learning model by:

obtaining a first batch of training data pairs comprising a given set of training radiotherapy treatment plan information;

processing the given set of training radiotherapy treatment plan information with the machine learning model to generate an intermediate radiotherapy treatment plan parameter;

computing a derivative of the loss function based on the intermediate radiotherapy treatment plan parameter; and

updating parameters of the machine learning model based on the computed derivative of the loss function.

15. A system comprising:

a memory for storing instructions; and

one or more processors for executing the instructions stored in the memory for performing operations comprising:

receiving radiotherapy treatment plan information;

processing the radiotherapy treatment plan information to estimate one or more radiotherapy treatment plan parameters based on a process that depends on an output of a subprocess that estimates a derivative of a dose calculation, wherein the derivative of the dose calculation is used in a machine learning model that is based on a loss function, and wherein the derivative of the dose calculation is computed with respect to at least one of one or more radiation parameters or one or more geometry parameters of a radiotherapy treatment device, wherein processing the radiotherapy treatment plan information comprises processing the radiotherapy treatment plan information with the machine learning model to generate the one or more estimated radiotherapy treatment plan parameters, the machine learning model being trained, based on a plurality of training dose calculations, to establish a relationship between a plurality of training radiotherapy treatment plan information and a plurality of training radiotherapy treatment plan parameters; and

generating a radiotherapy treatment plan using the estimated one or more radiotherapy treatment plan parameters.

16. The system of claim 15 , wherein the radiotherapy treatment plan information includes at least one of a magnetic resonance (MR) image, a cone-beam computed tomography (CBCT) image, a computed tomography (CT) image, a dose distribution, a segmentation map or a distance map.

17. The system of claim 16 , wherein the machine learning model comprises a deep neural network that is trained based on the plurality of training dose calculations using the plurality of training radiotherapy treatment plan information, wherein the plurality of training radiotherapy treatment plan information comprises at least one of a training magnetic resonance (MR) image, a training cone-beam computed tomography (CBCT) image, a training computed tomography (CT) image, a first training dose distribution, a training segmentation map or a training distance map, and wherein the plurality of training radiotherapy treatment plan parameters comprises at least one of a training synthetic computed tomography (sCT) image or a second training dose distribution.

18. The system of claim 15 , wherein the operations further comprise training the machine learning model by:

obtaining a first batch of training data pairs comprising a given set of training radiotherapy treatment plan information;

processing the given set of training radiotherapy treatment plan information with the machine learning model to generate an intermediate radiotherapy treatment plan parameter;

computing a derivative of the loss function based on the intermediate radiotherapy treatment plan parameter; and

updating parameters of the machine learning model based on the computed derivative of the loss function.

19. The system of claim 18 , wherein the first batch of training data pairs comprises a set of corresponding training radiotherapy treatment plan parameters associated with the given set of training radiotherapy treatment plan information, and wherein the operations further comprise:

computing a first dose based on the set of corresponding training radiotherapy treatment plan parameters;

computing a second dose based on the intermediate radiotherapy treatment plan parameter; and

applying the first and second doses to the loss function before computing the derivative of the loss function.