Three-dimensional dose prediction method based on deep learning and prior plan

Abstract

The invention discloses a three-dimensional dose prediction method based on deep learning and prior plan. The method includes obtaining comprehensive treatment-related information such as medical modal image, three-dimensional dose distribution information, target region structure images and normal organ structure images by collecting and processing data of existing high-quality radiation therapyplan, and inputting the above information into a pre-built neural network model for repeated training and optimization. The neural network model can fully learn the dose distribution of a high-qualityprior plan, and the resulting neural network model can more accurately predict the three-dimensional dose. Thus, the efficiency and effectiveness of the radiotherapy plan output by the model can be greatly improved, and a good generalization ability is gained.

Description

technical field [0001] The invention belongs to the field of radiotherapy plan design, and in particular relates to a three-dimensional dose prediction method based on deep learning and prior planning. Background technique [0002] Radiation therapy is one of the three main means of tumor treatment (surgery, radiotherapy and chemotherapy). With the development of precise radiation therapy technology, especially the popularization of complex treatment technology such as intensity-modulated radiotherapy (IMRT), it can also greatly reduce the radiation dose of normal tissues while meeting the dose coverage of the target area. radiation dose. However, complex treatment plans are not automatically generated by machines. Physicians first need to give the prescribed dose of the target area and the tolerance dose limit of different organs at risk, and then the physicist designs high-quality treatment plans based on various factors such as the target area and the structure of surrou...

AI Technical Summary

Problems solved by technology

[0003] The existing design-assisted radiotherapy plan design technology through modeling predicts the dose-volume histogram according to the distance relationship between the target volume and the organ at risk, such as the overlap volume (overlap volume histogram, OV), or based on the existing similar treatment plan. Three-dimensional dose is used as a template, and automatic planning and design are performed based on template prediction. However, these existing methods have the disadvantages of insufficient prediction accuracy and the need for a large amount of manpower to adjust the model. At the same time, the selected channels have less information and poor generalization ability.

Images