Three-dimensional dose distribution prediction method based on adaptive correction adversarial network

Abstract

The invention relates to a three-dimensional dose distribution prediction method based on an adaptive correction adversarial network, and provides a three-dimensional dose distribution prediction model based on the adaptive correction adversarial network by taking a generative adversarial network as a basic architecture and utilizing an ensemble learning idea. A dose distribution map as close as possible to target dose distribution is generated by using the idea of mutual game between a generator and a discriminator in the generative adversarial network. At the same time, the adaptive correction network is trained to fit a residual plot between generated and real dose distribution maps. The dose distribution map synthesized by the generator and the residual plot obtained by the adaptive correction network are superposed to obtain more ideal corrected dose distribution conforming to individual differentiation. According to the method, mapping between dosimetry characteristics and a geometric anatomical structure can be effectively learnt, and the network has higher robustness.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a three-dimensional dose distribution prediction method based on an adaptive correction confrontation network. Background technique [0002] Radiation therapy (referred to as "radiotherapy") is one of the mainstream methods in the treatment of malignant tumor patients. Whether it is radical radiotherapy or palliative radiotherapy, the fundamental purpose is to minimize the dose to surrounding organs at risk (organ at risk, OAR) while giving a curative dose to the tumor area. Therefore, increasing the dose ratio between tumor and OAR has become a key step to improve the patient's treatment gain ratio (ie, the ratio of tumor control rate to normal tissue complication rate). However, there are two problems in the quality of the radiotherapy plan: (1) The quality of the radiotherapy plan mainly depends on the knowledge level and subjective experience accumulation of the radiotherapy ...

AI Technical Summary

Problems solved by technology

However, there are two problems in the quality of the radiotherapy plan: (1) The quality of the radiotherapy plan mainly depends on the knowledge level and subjective experience accumulation of the radiotherapy physicists. The quality of radiotherapy plans is difficult to guarantee

(2) The clinical radiotherapy plan is restricted by uniform normative standards, which leads to the fact that the plan design meets the clinical norms, but ignores the individual differences among patients

However, in the dose distribution prediction task, the input image is the geometric anatomical structure map of the patient, and the predicted image is the dose distribution map. The feature space distribution of the two is quite different, and the U-shaped network and its transformation and the traditional generation confrontation network are purely relied on. The association between dosimetric features and their geometrical anatomy is often not accurately learned

Images