MACHINE LEARNING APPROACH FOR SOLVING BEAM ANGLE OPTIMIZATION
Embodiments described herein provide for revising radiation therapy treatment plans, and in particular, revising beam angles used during radiation therapy treatment. A computer may receive a radiation therapy treatment plan based on a particular patient's diagnosis. The computer may use a machine learning model to revise radiation therapy treatment parameters such as a beam angle indicating a direction of radiation into the patient. The machine learning model may use reinforcement learning to optimize an initial beam angle from the radiation therapy treatment plan, revising the beam angle. The performance of the machine learning model is measured against metrics including fulfilling dosimetric clinical goals. The machine learning model may present the revised beam angle for display to a medical professional, or transmit the revised beam angle to downstream applications to further revise the radiation therapy treatment plan.
1 . A computer-implemented method of beam angle optimization comprising:
executing, by at least one processor, a machine learning model that receives an input of data associated with a treatment plan for a patient and outputs a beam angle for the patient indicating a direction of radiation into the patient,
wherein the machine learning model is trained using a training dataset comprising a training treatment plan and a corresponding score,
wherein the machine learning model iteratively calculates a reward, using a policy, for a possible beam angle for the training treatment plan in the training dataset, and
wherein the machine learning model iteratively increases a summation of rewards until the policy satisfies an accuracy threshold; and
transmitting, by the at least one processor, the beam angle to a second processor.
2 . The computer-implemented method according to claim 1 , wherein at least one of the treatment plan or the training treatment plan comprise at least one of a medical image, a clinical goal, a planning target volume, an organ at risk, a radiation type, a radiation dose, an initial beam angle, or a field geometry.
3 . The computer-implemented method according to claim 2 , wherein the medical image includes at least a structure of the planning target volume or a structure of the organ at risk.
4 . The computer-implemented method according to claim 1 , further comprising executing the machine learning model that receives the input of data associated with the treatment plan for the patient and outputs a dose distribution, wherein the machine learning model is trained using a training dataset comprising the training treatment plan and a corresponding score.
5 . The computer-implemented method according to claim 1 , further comprising presenting, by the processor, for display, the beam angle.
6 . The computer-implemented method according to claim 1 , wherein the machine learning model is trained using asynchronous advantage actor critic reinforcement learning.
7 . The computer-implemented method according to claim 1 , wherein the machine learning model is implemented using hybrid graphics processing units and central processing units.
8 . The computer-implemented method according to claim 1 , wherein the machine learning model is optimized with respect to one or more clinical goals received in the treatment plan, the clinical goals including at least one of a dosimetric quality, a robustness measure, metrics based on linear energy transfer, or relative biological effects.
9 . The computer-implemented method according to claim 1 , further comprising:
receiving, by the at least one processor from the second processor, a revised treatment plan, wherein the revised treatment plan is based on the beam angle;
executing, by the at least one processor, the machine learning model using the revised treatment plan for the patient and outputting a revised beam angle; and
transmitting, by the at least one processor, the revised beam angle to the second processor.
10 . The computer-implemented method according to claim 1 , wherein iteratively calculating the reward, using the policy, for the possible beam angle from the training treatment plan in the training dataset includes iteratively comparing the reward to a baseline.
11 . A system comprising:
a server comprising a processor and a non-transitory computer-readable medium containing instructions that when executed by the processor causes the processor to perform operations comprising:
execute a machine learning model that receives an input of data associated with a treatment plan for a patient and outputs a beam angle for the patient indicating a direction of radiation into the patient, wherein the machine learning model is trained using a training dataset comprising a training treatment plan and a score, wherein the machine learning model iteratively calculates a reward, using a policy, for a possible beam angle for the training treatment plan in the training dataset, wherein the machine learning model iteratively increases a summation of rewards until the policy satisfies an accuracy threshold; and
transmit the beam angle to a second processor.
12 . The system according to claim 11 , wherein at least one of the treatment plan or the training treatment plan comprise at least one of a medical image, a clinical goal, a planning target volume, an organ at risk, a radiation type, a radiation dose, an initial beam angle, or a field geometry.
13 . The system according to claim 12 , wherein the medical image includes at least a structure of the planning target volume or a structure of the organ at risk.
14 . The system according to claim 11 , wherein the processor is further configured to execute the machine learning model that receives the input of data associated with the treatment plan for the patient and outputs a dose distribution, wherein the machine learning model is trained using a training dataset comprising the training treatment plan and a corresponding score.
15 . The system according to claim 11 , wherein the processor is further configured to present for display, the beam angle.
16 . The system according to claim 11 , wherein the machine learning model is trained using asynchronous advantage actor critic reinforcement learning.
17 . The system according to claim 11 , wherein the machine learning model is implemented using hybrid graphics processing units and central processing units.
18 . The system according to claim 11 , wherein the machine learning model is optimized with respect to one or more clinical goals received in the treatment plan, the clinical goals including at least one of a dosimetric quality, a robustness measure, metrics based on linear energy transfer, or relative biological effects.
19 . The system according to claim 11 , wherein the processor is further configured to:
receive, from the second processor, a revised treatment plan, wherein the revised treatment plan is based on the beam angle;
execute the machine learning model using the revised treatment plan for the patient and outputting a revised beam angle; and
transmit the revised beam angle to the second processor.
20 . The system according to claim 11 , wherein iteratively calculating the reward, using the policy, for the possible beam angle from the training treatment plan in the training dataset includes iteratively comparing the reward to a baseline.