Method and apparatus for controlling multileaf collimator in radiotherapy

1. A method for controlling multileaf collimator in radiotherapy, comprising:

acquiring a preset model;

outputting, regarding a radiotherapy target, a target leaf sequence of a multileaf collimator through the preset model, to obtain a planning file; and

importing the planning file into a radiotherapy planning system to control a linear accelerator to operate, so as to implement a radiotherapy plan;

wherein before acquiring a preset model, the method further comprises:

acquiring sample data, wherein the sample data comprises: fluence maps in historical planning files; and

performing reinforcement learning and training through a neural network model by using the sample data, to generate the preset model.

2. The method as claimed in claim 1 , wherein before acquiring sample data, the method further comprises:

setting environmental parameters of a training frame according to a model of the linear accelerator, wherein the environmental parameters at least comprise: a width of each leaf, a movement range of each leaf, a movement speed of each leaf, and a total number of leaves of the multileaf collimator;

setting states of leaves of the training frame, wherein the states at least comprise: a leaf position where a current leaf is located, a currently obtained fluence map, and a target fluence map; and

setting a reward function for the leaves of the training frame, wherein the reward function is used for calculating a difference between the currently outputted fluence map and the target fluence map, and allocating a reward corresponding to the leaf position.

3. The method as claimed in claim 2 , wherein before acquiring sample data, the method further comprises:

determining a simulated state of the multileaf collimator in an environment according to the environmental parameters, the states and the reward function.

4. The method as claimed in claim 3 , wherein performing reinforcement learning and training through a neural network model by using the sample data to generate the preset model, comprises:

performing reinforcement learning and training on the sample data and the simulated state of the multileaf collimator by a first network of the neural network model to obtain a simulated leaf sequence of the multileaf collimator, wherein the simulated leaf sequence of the multileaf collimator comprises: coordinate positions of the leaves and weights of irradiation fields;

evaluating the simulated leaf sequence of the multileaf collimator by a second network in the neural network model; and

adjusting parameters in the first network and the second network of the neural network model according to an evaluation result to generate the preset model.

5. The method as claimed in claim 1 , wherein after importing the planning file into a radiotherapy planning system to control a linear accelerator to operate, so as to implement a radiotherapy plan, the method further comprises:

importing the planning file into the radiotherapy planning system to control the linear accelerator to operate, to acquire an actual fluence map; and

evaluating the preset model based on the actual fluence map and the target fluence map.

6. A non-transitory storage medium, wherein the non-transitory storage medium comprises a program stored therein, wherein when the program runs, a device where the non-transitory storage medium is located is controlled to perform the method as claimed in claim 1 .

7. An electronic apparatus, comprising a processor and a memory; wherein computer-readable instructions are stored in the memory, and the processor is configured to run the computer-readable instructions, wherein when the computer-readable instructions are running, the method as claimed in claim 1 is implemented.

8. An apparatus for controlling multileaf collimator in radiotherapy, comprising:

a first acquisition unit, which is configured to acquire a preset model;

a second acquisition unit, which is configured to output, regarding a radiotherapy target, a target leaf sequence of a multileaf collimator through the preset model, to obtain a planning file; and

a control unit, which is configured to import the planning file into a radiotherapy planning system to control a linear accelerator to operate, so as to implement a radiotherapy plan;

wherein the apparatus further comprises:

a third acquisition unit, which is configured to acquire sample data before acquiring a preset model, wherein the sample data comprises: fluence maps in historical planning files; and

a generation unit, which is configured to perform reinforcement learning and training through a neural network model by using the sample data, to generate the preset model.