Quantification and classification of Magnetic Resonance Spectroscopy data and applications to brain tumour recognition

The medical diagnosis of brain tumours is one of the main applications of Magnetic Resonance. Magnetic Resonance consists of two main branches: Imaging and Spectroscopy. Magnetic Resonance Imaging is very well-known as the radiologic technique applied to produce high-quality images of tissues, such as the brain tissue, for diagnostic purposes. Magnetic Resonance Spectroscopy provides chemical information about all the molecules present in the brain, such as their concentrations. Both Imaging and Spectroscopy can be exploited for the grading and typing of brain tumours, also called the classification of brain tumours. As first topic, this thesis mainly studied the contribution of Spectroscopy for automated classification and the influence of several factors on the classification performance. It was found that a few preprocessing steps did not have a large impact on the classification results. This implies that several preprocessing steps can be ignored for diagnostic purposes and that acquisition schemes can be simplified. Furthermore, classification can be based on a restricted number of extracted parameters that cover the most characteristic information, which simplifies the computation and reduces the computation time. In addition, it was observed that spectroscopic and imaging parameters can provide complementary information for the typing of brain tumours. A second topic was the processing of experimental glycogen signals, which provides a better understanding of the biochemical processes involved in diseases such as diabetes. However, the spectral processing needs improvement. A common problem is that most methods assume the number of components present to be known exactly, but in practice this knowledge is lacking. Especially for so-called multi-exponential signals, in which different components occur at the same spectral frequency but with different linewidths. A framework was applied that determined the number of components for glycogen signals and that was able to confirm the presence of a multi-exponential pattern in glycogen signals.