Nieuwe kernmerkextractie methoden voor de verbeterde analyse van magnetische resonantie spectroscopie en spectroscopische beeldvorming.

Nuclear magnetic resonance spectroscopy (MRS) has an enormous potential to contribute to better understanding and improved diagnosis and prognosis for various diseases. Due to the high complexity of the MRS data, for this technique to be clinically successfully accepted, reliable and robust automatic (pre)processing and classification strategies are needed. With this project, we aim to facilitate the interpretation and analysis of in vivo and ex vivo MRS data, by proposing advanced feature extraction and selection methods, which will improve classification and diagnosis. The challenging problem of accurate metabolite quantification in 2D and 3D data sets is addressed by using model based quantification methods which incorporate spatial information. The proposed methods will be applied for cancer diagnosis. Another application for improved classification of MRS data is the identification of disease specific biomarkers in amniotic fluid. A semi-supervised optimal feature selection method in combination with automatic classification is proposed. This represents a pioneering approach with high potential for a deeper understanding of the pathophysiology of several clinical conditions in fetuses.

Keywords:Magnetic resonance spectroscopy, Feature extraction, Metabolite quantification, Disease biomarkers

Disciplines:Modelling, Multimedia processing