SYSTEM AND METHOD FOR LEARNING MODELS OF PLANS OF RADIOTHERAPEUTIC TREATMENT WITH PREDICTION OF DOSE DISTRIBUTION OF RADIOTHERAPY

摘要

FIELD: medicine.;SUBSTANCE: invention refers to medicine, namely to medical equipment. A radiation therapy prediction system with a radiation dose prediction comprises: an image acquisition device for obtaining one or more three-dimensional medical images; non-temporal machine-readable data medium for storing said one or more three-dimensional medical images, a neural network model, one or more three-dimensional anatomical maps and one or more three-dimensional dose distributions; image processing device configured to: teach a neural network model to predict at least one of the fluence map and the dose map based on said one or more three-dimensional medical images and said one or more three-dimensional anatomical maps; and generate a three-dimensional dose distribution based on neural network prediction. A radiation therapy prediction system with a radiation dose prediction comprises: an image acquisition device for obtaining a set of training data, wherein the training data comprise one or more three-dimensional medical images, a neural network model, one or more three-dimensional anatomical maps and one or more three-dimensional dose distributions, a non-temporal machine-readable data medium for storing training data, a first neural network model and a second neural network model; wherein the image processing device is configured to: teach the first neural network model using the training data to predict the first dose distribution; train the second neural network model using the training data to predict the second dose distribution; determining an error by comparing a first dose distribution with a second dose distribution; and use this error to train the first neural network model. Method for prediction of a dose of radiation therapy comprises: receiving one or more three-dimensional medical images from an image acquisition device; storing three-dimensional images, neural network models, one or more three-dimensional anatomical maps and one or more three-dimensional dose distributions on the non-time machine-readable data medium; training the neural network through at least one processor to predict at least one of the fluence map and the dose map based on said one or more three-dimensional medical images and said one or more three-dimensional anatomical charts and one or more three-dimensional dose distributions; and generating a three-dimensional dose distribution based on neural network prediction.;EFFECT: technical result of the declared invention consists in improvement of accuracy in optimization of the treatment plan.;22 cl, 19 dwg