Challenging the state-of-the-art of in vivo brain MRSI

The human brain is a hardly understood organ with a complex structure, function and metabolism. The metabolism in the brain is altered in the case of several diseases such as brain tumours, multiple sclerosis, or epilepsy. This change in metabolism occurs before any structural or functional changes. Therefore, if it is possible to measure the brain metabolism, these diseases can potentially be detected in their very early stages, when treatment is also usually more effective. Magnetic Resonance Spectroscopic Imaging (MRSI) is such a methodology to investigate the metabolism in the human body using an MR scanner. Such investigations do not use ionizing radiation or radioactive substances, and are therefore not invasive or harmful. With this technique we can measure chemicals in the brain like creatine, choline, glutamate, or N-acetyl-aspartate, which have been shown to change in brain diseases in many studies. Unfortunately, the concentrations of these chemicals are magnitudeslower than water and MR- visible lipids. Thus, these water and lipid signals overwhelm the signals of the other chemicals and sometimes makes it even impossible to quantify them. As a result, great effort has been made to suppress the water and lipid signals during acquisition, mitigate their effects by appropriate design of the experiments, or remove them in post-processing. However, each group is having their own methods and approaches, which are hardly shared between them and are hard to compare and it is difficult to evaluate which method is better. Therefore, the goal of this project is to conduct an international challenge about the removal of such artefactsignals, where groups from all over the world can participate and try to remove these artefact signals from data prepared by us. This data preparation will include the simulation of very realistic MRSI data, mimicking all known physical effects happening at an MR scanner during acquisition. By simulating instead of measuring the data we will get ground truth signals without artefacts, together with the data containing artefacts, so that we can compare the contributions of the challenge participants against the known artefact-free ground truth. The data with artefacts will then be made available online for all participants to be used for their artefact removal methods. After gathering the results of all groups, the best contributions will be awarded in several benchmark categories. With this project we hope to foster collaboration and sharing of source code between groups, and we hope to make the different artefact removal methods more comparable and determine which works better for certain data. We also want to establish a benchmark so that future methods can easily be compared against the already established methods. Furthermore, by performing very realistic simulations, we will be able to provide training data for machine learning algorithms, as these methods require big and realistic datasets.