Automatic radiotherapy dose prediction method based on deep neural network

Abstract

The present invention relates to the field of dose prediction in radiotherapy, and proposes an automatic radiotherapy dose prediction method based on a deep neural network, comprising the following steps: firstly, obtaining a CT image of a patient; secondly, delineating a patient's target area and organs at risk based on the CT image; Then, calculate the non-modulated dose of the patient, where the non-modulated dose is used to provide global information; then, build a dose prediction model based on deep neural network based on CT images, target volumes, organs at risk and non-modulated dose; finally, use depth-based The dose prediction model of neural network can automatically predict the radiotherapy dose. The present invention can automatically predict patient dose distribution based on deep neural network, the process does not need to manually extract features or set parameters, and the predicted dose distribution can accelerate the formulation of radiotherapy plan for patients, and the model uses non-modulated dose in the input part , which can provide more global information for accurate prediction of dose distribution.

Description

technical field [0001] The invention relates to the field of dose prediction in radiotherapy, in particular to an automatic radiotherapy dose prediction method based on a deep neural network. Background technique [0002] Radiation therapy, referred to as radiotherapy, is one of the main methods of tumor treatment. Its working principle is to irradiate tumors with high-energy radiation, destroying the DNA structure of tumors, so as to kill tumor cells and protect non-tumor tissues. Radiotherapy has the advantages of wide application range, few side effects, low trauma and painless, etc. It can effectively improve the cure rate of patients and prolong the life of patients. [0003] In clinical applications, before the implementation of radiotherapy, it is necessary for professionals to formulate a radiotherapy plan, that is, according to the target area of ​​the patient and the organs at risk drawn by the doctor, combined with the prescribed dose, optimize the angle and inten...

AI Technical Summary

Problems solved by technology

[0006] However, it is noted that the target area and organs at risk included in the input can only provide the "local" structural information of the patient's anatomical structure, but the output is the "global" dose distribution of the patient, learning the mapping relationship from "local" to "global" to the depth Neural Network Models Present Challenges

In addition, the input data does not contain information about the X-ray penetration area, which increases the learning difficulty of the model

[0007] Therefore, the objective shortcomings of the prior art: existing work studies the relationship between the target area and organs at risk and the DVH curve, and the DVH curve only contains dose information in a statistical sense, which cannot meet the requirements for dose distribution in clinical applications

In the work of using deep neural network to predict patient dose, the input data failed to provide "global" information related to dose, and the input data did not include information about the X-ray penetration area

Images