This is a class project for CIS 522 Deep Learning (Spring 2020) at University of Pennsylvania.
Abstract
High-field MRI offers superior resolution and contrast compared to lower field strength MRI. However, high-field MRI is not as easily available and more costly. Therefore, a computational method for increasing resolution and contrast in medical imaging would be of significant medical and scientific importance. Here, we show that a generative adversarial network can be trained to transform 7T images to look more like 9.4T images by improving the contrast and resolution of the original images. We also show that this method performs superior to a fully convolutional network without a paired discriminator as well as a non-deep learning random forest model.
See our full report here!