End-to-End Quantitative Susceptibility Mapping

Increased brain iron accumulation is a common finding in neurological disorders such as Alzheimers disease, Parkinsons disease, or multiple sclerosis. Because of its paramagnetic nature, iron is changing the magnetic susceptibility of brain tissue and recent validation studies showed that brain iron can be measured precisely by the novel magnetic resonance imaging (MRI) technique quantitative susceptibility mapping (QSM) in vivo, thus, enabling reliable and precise longitudinal investigations in neurological disorders. In this work we utilize machine learning techniques to solve the mathematical problem of calculating QSM from a series of images from an MRI scanner. In contrast to conventional techniques, machine learning uses artificial neural networks which are trained using dedicated hardware. QSM is a multistep approach where each step modifies the MRI images from the scanner in a certain way, and numerical errors propagate for each step. In this project we will combine those individual steps in a single artificial neural network so that the raw images from the MRI system can be used directly and the overall calculation error is minimized. Additionally, free parameters are not necessary for the machine learning algorithm which allows better comparability as well as the implementation directly on a clinical MRI scanner.