Enhancing MRI diagnosis of myocarditis using deep learning and generative adversarial networks

Abstract

In this paper, in order to enhance the MRI diagnosis of myocarditis, a generative adversarial network (GAN)-based MRI diagnostic model for myocarditis is constructed in this paper. The MRI images of myocarditis provided by a hospital were used as the data source for this study, and the image format was transformed into NII format file for saving using Python tool, which was uniformly cropped to 480×768 pixels, and stored in the form of datasets, which were divided into dataset A (the MRI-weighted images of the myocarditis dataset) and dataset B (the MRI images of myocarditis). ResNet-34 network and U-Net network were used as the generator and discriminator, respectively, and in order to address the problem of difficulty in training GAN networks, a BN layer was added between the convolutional layer and the activation function in the generator and the discriminator, and the construction of the model was finally completed. Determine the loss function, select the quantitative evaluation indexes of the model (MAE, RMSE, PSNR, SSIM and PCC), set the control model (CNN, RNN, LSTM, GRU), and validate and analyze the model in this paper. The generator loss function and discriminator loss function after 400 iterations of training, the value of the loss of both is almost 0. The quantitative evaluation indexes of this paper’s model genus pig are higher than the other four models. In summary, generative adversarial network has a facilitating effect on MRI diagnosis of myocarditis.