Automatic plan optimization system based on coupled generative adversarial network

Abstract

The invention discloses an automatic plan optimization system based on a coupled generative adversarial network, and the system comprises an information input module which is used for inputting required data information; a weight automatic setting module which is used for automatically setting the weight of a sub-objective function through an objective function weight factor prediction model based on the convolutional neural network; a constraint condition setting module which is used for automatically setting constraint conditions of the sub-objective functions through a dosimetry indication item prediction model based on a BP neural network; a plan optimization module which is used for training by utilizing a generative adversarial network to obtain a sub-field shape prediction model, optimizing an objective function by adopting a local gradient algorithm according to the content to obtain optimal dose distribution matrix data, and generating an optimal plan; and a scheme output module which is used for outputting a plan scheme according to the optimal plan generated by the plan optimization module. The intensity modulated radiotherapy plan is optimized by using the convolutional neural network, the BP neural network and the generative adversarial network, so that the optimization process is more efficient, and the quality of the optimization result is improved.

Description

technical field [0001] The invention relates to the technical field of intensity-modulated radiotherapy plan optimization, in particular to an automatic plan optimization system based on a coupled generative confrontation network. Background technique [0002] Intensity modulated radiation therapy (IMRT) is one of the mainstream technologies in the current clinical treatment of cancer. Its plan design adopts the method of reverse plan design. First, the doctor determines the target PTV and related organs at risk on the patient's CT image. OARs information, and then the physicist uses this information to set the optimization target constraints and weight factors of PTV and OARs, and finally adjusts the intensity of each photon beam through the planning optimization algorithm to achieve the three-dimensional conformity of the dose in the PTV. Concentrating the radiation dose in the PTV to the greatest extent and avoiding unnecessary exposure to surrounding OARs and normal tiss...

AI Technical Summary

Problems solved by technology

However, in high-dimensional problems with many PTVs and OARs, almost all individuals become independent of each other, the search ability of the algorithm will decline sharply, and it is impossible to find a Pareto optimal solution or an approximate Pareto optimal solution, and the resulting treatment plan is not optimal plan

The KBRT algorithm, on the one hand, requires a large amount of clinical data as input to determine the weight value of the current plan optimization, and the process is very cumbersome; on the other hand, due to the large individual differences between patients, especially for patients with complex anatomical structures Cancer, resulting in the prediction of the weight values ​​required for each patient plan optimization is not accurate enough to obtain the optimal plan

Automatic optimization algorithm, this type of algorithm usually needs multiple iterations according to the plan optimization results, and for each set of selected weight factors, the dose distribution needs to be re-optimized, which requires a large amount of calculation and takes a long time

Images