Signal processing and classification for magnetic resonance spectroscopic data with clinical applications

Over the last decades, Magnetic Resonance Imaging (MRI) has taken a leading role in the study of human body and it is widely used in clinical diagnosis. In vivo and ex vivo Magnetic Resonance Spectroscopic (MRS) techniques can additionally provide valuable metabolic information as compared to MRI and are gaining more clinical interest. The analysis of MRS data is a complex procedure and requires several preprocessing steps aiming to improve the quality of the data and to extract the most relevant features before any classification algorithm can be successfully applied. In this thesis a new approach to quantify magnetic resonance spectroscopic imaging (MRSI) data and therefore to obtain improved metabolite estimates is proposed. Then an important part is focusing on improving the diagnosis of glial brain tumors which are characterized by an extensive heterogeneity since various intramural histopathological properties such as viable tumor cells, necrotic tissue and infiltration with normal tissue can be identified in the tumor region. For a reliable diagnosis of the glial tumor type and grade this thesis proposes a first screening between these intratumoral histopathological properties. To this aim, cluster analysis and several blind source separation methods are tested on ex vivo HR-MAS and in vivo MRSI data. Moreover, several approaches to fuse multimodal information coming from MRI, MRSI and HR-MAS for the classification of glial brain tumors are considered. MRS techniques are nowadays successfully considered for the analysis of body fluids. A pilot research to study the amniotic fluid from fetuses with congential diaphragmatic hernia using high resultion MRS spectroscopy is proposed.