Neural network calibration for radiotherapy

1. A method comprising:

accessing, by one or more processors, a neural network trained based on a first set of data generated from characteristic values of a first set of patients that received treatment at a set of one or more first radiotherapy machines;

executing, by the one or more processors, the neural network using a second set of data comprising characteristic values of a second set of patients receiving treatment at a set of one or more second radiotherapy machines to output a set of treatment attribute predictions, the second set of data having corresponding labels indicating expected treatment attribute predictions;

executing, by the one or more processors, a calibration model using the set of treatment attribute predictions and labels indicating expected treatment attribute predictions to output a calibration value;

executing, by the one or more processors, the neural network using a set of characteristics of a first patient of the second set of patients receiving treatment at the set of one or more second radiotherapy machines to output a first confidence score associated with a first treatment attribute; and

adjusting, by the one or more processors, the first confidence score according to the calibration value to predict the first treatment attribute.

2. The method of claim 1 , further comprising:

executing, by the one or more processors, the neural network using a set of characteristics of a second patient of the second set of patients receiving treatment at the set of one or more second radiotherapy machines to output a second confidence score associated with a second treatment attribute and adjusting the second confidence score according to the calibration value to predict the second treatment attribute.

3. The method of claim 1 , wherein predicting the first treatment attribute comprises comparing the adjusted first confidence score to a threshold and predicting the first treatment attribute responsive to determining the adjusted first confidence score exceeds the threshold.

4. The method of claim 1 , further comprising:

adjusting, by the one or more processors, a configuration of a second radiotherapy machine of the set of one or more second radiotherapy machines according to the first treatment attribute.

5. The method of claim 1 , further comprising:

rendering, by the one or more processors, the adjusted first confidence score and the first treatment attribute on a display;

receiving, by the one or more processors, a user input selecting the first treatment attribute; and

responsive to receiving the user input, adjusting, by the one or more processors, a configuration of a second radiotherapy machine of the set of second one or more second radiotherapy machines based on the first treatment attribute.

6. The method of claim 1 , wherein the neural network is a first neural network and the calibration value is a first calibration value, further comprising:

executing, by the one or more processors, a second neural network using the set of characteristics of the first patient to output a second confidence score for a second treatment attribute; and

adjusting, by the one or more processors, the second confidence score associated with the second treatment attribute according to a second calibration value corresponding to the second neural network, the second calibration value generated based on the second set of data;

wherein predicting the first treatment attribute is performed responsive to the first confidence score exceeding the second confidence score.

7. The method of claim 1 , wherein the first treatment attribute is an angle associated with a couch or a gantry of a radiotherapy machine.

8. The method of claim 1 , wherein predicting the first treatment attribute comprises:

executing, by the one or more processors, a plurality of neural networks using the set of characteristics of the first patient to output confidence scores for a plurality of treatment attributes, each neural network associated with a different calibration value;

adjusting, by the one or more processors, the output confidence scores of each of the plurality of neural networks according to the calibration value associated with the respective neural network; and

predicting, by the one or more processors, the first treatment attribute by aggregating the adjusted confidence scores that correspond to individual treatment attributes.

9. The method of claim 1 , further comprising:

adjusting, by the one or more processors, the calibration value responsive to training the neural network according to a supervised learning algorithm from a training set of values labeled according to whether a radiotherapy machine adjusted its configuration based on a predicted treatment attribute.

10. The method of claim 1 , wherein the calibration model is one of a machine learning model or a mathematical optimization algorithm.

11. The method of claim 1 , further comprises:

predicting, by the one or more processors, a third confidence score associated with a third treatment attribute by executing the neural network using a set of characteristics of a third patient receiving treatment at the second radiotherapy clinic;

adjusting, by the one or more processors, the third confidence score according to the calibration value; and

responsive to determining the adjusted third confidence score does not exceed a threshold, rendering, by the one or more processors, the third confidence score and an indication that the third confidence score does not exceed the threshold on a display.

12. The method of claim 1 , wherein the neural network is trained based only on data associated with the first set of patients.

13. A system comprising:

one or more processors in communication with a radiotherapy machine, the processor configured to execute instructions to:

access a neural network trained based on a first set of data generated from characteristic values of a first set of patients that received treatment at a set of one or more first radiotherapy machines;

execute the neural network using a second set of data comprising characteristic values of a second set of patients receiving treatment at a set of one or more second radiotherapy machines to output a set of treatment attribute predictions, the second set of data having corresponding labels indicating expected treatment attribute predictions;

execute a calibration model using the set of treatment attribute predictions and labels indicating expected treatment attribute predictions to output a calibration value;

execute the neural network using a set of characteristics of a first patient of the second set of patients receiving treatment at the set of one or more second radiotherapy machines to output a first confidence score associated with a first treatment attribute; and

adjust the first confidence score according to the calibration value to predict the first treatment attribute.

14. The system of claim 13 , wherein the processor is further configured to:

execute the neural network using a set of characteristics of a second patient of the second set of patients receiving treatment at the set of one or more second radiotherapy machines to output a second confidence score associated with a second treatment attribute and adjusting the second confidence score according to the calibration value to predict the second treatment attribute.

15. The system of claim 13 , wherein the processor is configured to predict the first treatment attribute by comparing the adjusted first confidence score to a threshold and predicting the first treatment attribute responsive to determining the adjusted first confidence score exceeds the threshold.

16. The system of claim 13 , wherein the processor is further configured to:

adjust configuration of a second radiotherapy machine of the set of one or more second radiotherapy machines according to the first treatment attribute.

17. The system of claim 13 , wherein the processor is further configured to:

render the adjusted first confidence score and the first treatment attribute on a display;

receive a user input selecting the first treatment attribute; and

responsive to receiving the user input, adjust a configuration of a second radiotherapy machine of the set of second one or more second radiotherapy machines based on the first treatment attribute.

18. The system of claim 13 , wherein the neural network is a first neural network and the calibration value is a first calibration value, and wherein the processor is further configured to:

execute a second neural network using the set of characteristics of the first patient to output a second confidence score for a second treatment attribute; and

adjust the second confidence score associated with the second treatment attribute according to a second calibration value corresponding to the second neural network,

wherein the processor is configured to predict the first treatment attribute responsive to the first confidence score exceeding the second confidence score.

19. The system of claim 13 , wherein the first treatment attribute is an angle associated with a couch or a gantry of a radiotherapy machine.

20. The system of claim 13 , wherein the processor is configured to predict the first treatment attribute by:

executing a plurality of neural networks using the set of characteristics of the first patient to output confidence scores for a plurality of treatment attributes, each neural network associated with a different calibration value;

adjusting the output confidence scores of each of the plurality of neural networks according to the calibration value associated with the respective neural network; and

predicting the first treatment attribute by aggregating the adjusted confidence scores that correspond to individual treatment attributes.