GAN architecture and method for performing data augmentation on medical image data set based on generative adversarial network

Abstract

The invention discloses a GAN architecture and method for performing data augmentation on a medical image data set based on a generative adversarial network (GAN). The method comprises the steps of: acquiring a real data set of an existing medical image; in the samples, taking out samples containing lesions and samples not containing lesions to serve as a group to be input together, and operatinga cyclic generative adversarial network to obtain artificial samples similar to real data; adding the artificial sample into the real data set to obtain a mixed data set; and taking the mixed data setas input, and performing a classification task by using a classifier. According to the method, a reconstruction consistency loss function constraint condition is introduced, source distribution is converted into target distribution, and then the source distribution is reconstructed; finally, a stable normalization layer is added in a discriminator, the distribution characteristics of real data are effectively simulated, an image is generated through the generative adversarial network for data enhancement, then a large number of medical image samples are simulated, and the influence of insufficient data samples on a medical image data classification task is effectively improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a network architecture and an image conversion method based on a generative confrontation network and a convolutional neural network. Background technique [0002] In recent years, with the rapid development of information technology and the continuous improvement of computer application level, deep learning technology and medical imaging technology, such as computed tomography (computed tomography, CT), positron emission computed tomography (positron emission computed tomography, PET) , single photon emission computed tomography (SPECT), magnetic resonance imaging (magnetic resonance imaging, MRI), ultrasound imaging, and images obtained by other medical imaging equipment are widely used in medical diagnosis, anatomical structure learning, treatment planning, and functional imaging. In various medical links such as data and postoperative monitoring. The rap...

AI Technical Summary

Problems solved by technology

Specifically, the present invention provides a sentiment classification method based on the method of generating confrontational network data. This method generates images by generating confrontational networks to make up for the problem of unbalanced data categories in the original, or the problem that the data set is very small. Data augmentation, and then train the classifier, thus improving the model accuracy

Images