Deep learning applications in magnetic resonance imaging

Magnetic resonance imaging (MRI) is a non-invasive and versatile imagining technique with applications ranging from tumor detection to brain functional imaging. Recent successes with Deep Learning algorithms, specifically convolutional neural networks (CNNs), in image classification and segmentation tasks have led to significant research in determining the capabilities of extending these algorithms to MRI. In this thesis, we explore CNN capabilities with two distinctly different tasks: classification of adipose tissue and brain functional connectivity analysis. For adipose tissue classification, we find that not only are CNNs capable of achieving state of the art accuracies based on pre-processed datasets, but these accuracies hold when used upon the raw complex signals derived from the MRI scanner. For brain functional connectivity analysis, we explore a new method for discovering resting state networks in pig brains. Pigs are a popular research replacement for human subjects, but brain functional connectivity is largely unexplored in these animals.