Student work

In our research group, we are developing new methods for magnetic resonance imaging (MRI) at ultrahigh magnetic fields (e.g., 7 Tesla). Our main goal is to be able to use the higher signal-to-noise ratio for in vivo human applications. However, a problem here is the spatially inhomogeneous distribution of the radio frequency (RF) field necessary to excite the nuclear spins that result in spatial variations of the image signal intensity. It has been shown that this spatial inhomogeneity can be successfully compensated by so-called multi-element RF coils: homogeneous signals and contrast can be generated by an optimized superposition of the transmitting fields of e.g. 8 independent RF transmitting elements.
However, an essential prerequisite for the optimization of the resulting field and the subsequent "parallel transmission" is that the RF fields of the individual elements are known. These must be measured for each subject at the beginning of the actual examination and this process takes too long with current measurement times of 5-15 minutes. Thus, it severely hinders the conductance and performance of patient studies at 7 Tesla.

Recently, a new method has been developed within the group, which can predict these RF fields within a few seconds using a neural network. As input of this network a fast localizer image is used, which is acquired anyway at the beginning of each measurement. These predicted maps can then be used to optimize the transmit fields to obtain a homogeneous image intensity in a two-dimensional layer.

The goal of this work is to systematically investigate and extend the accuracy of this method. This objective is divided into three work packages.

1. In the first step the method shall be systematically investigated for its accuracy compared to conventional mapping methods.
2. In the second step the network shall be extended so that it can also use 3D data instead of only 2D data so far. This will then allow the RF fields to be determined over a three-dimensional volume.
3. In the last step, the maps determined in this way will be used to calculate RF pulses that achieve homogeneous excitation of a 3D volume. These pulses are to be implemented in a suitable MR sequence and finally validated in a study with human subjects.

• Study of physics, electrical engineering, computer science, biomedical engineering or a comparable course of studdy
• good programming skills (Python/MATLAB)
• initial experience in the field of deep learning is advantageous
• interest in MR physics
• You are a strong team player with excellent communication skills

Dr. Sebastian Schmitter, Working Group 8.14
E-Mail: sebastian.schmitter(at)ptb.de
Tel.: + 49 30 3481-7767