Magnetic Resonance Spectroscopy based diagnostic method development

OData support
Dr. Benyó Balázs István
Department of Control Engineering and Information Technology

MR spectra (MRS) can be used to derive metabolite concentrations of a localized region in vivo in a non invasive way.

Following data acquisition of a time domain signal, it undergoes preprocessing steps. These steps are to correct for artefacts causing lineshape distortions and phase or frequency shifts and to remove residual water or other unwanted components. Severe signal distortions most often appear in biomedical MRS due to instrumental imperfections and sample inhomogeneities.

In the processing step a wide range of methods can be used to perform quantification. Theoretically time- and frequency domain analysis is equivalent, however the latter is computationally more demanding. Usually, visualization is performed in the frequency region. Time domain signal is fitted by a sum of exponentially decaying sinusoids. Iterative methods allow the inclusion of constraints on frequency, damping, phase and amplitude parameters. Prior knowledge can be implicitely imposed by in-vitro measured metabolite profiles or simulated time domain signals. In this way in vivo signal can be fitted as the linear combination of metabolite profiles. The parameter of interest are the amplitudes (weights) of each metabolite as it is proportional to the concentration. A baseline is reconstructed via penalized splines in a non parametric optimization step. Several parameter estimation method exist to determine the adjustments. AQSES performs a regularized non-linear least squares algorithm using variable projection. This method is preferred when processing short echo time signals. The most important drawbacks are the required user interaction.

My work has focused on refining the consecutive preprocessing steps in a template so that large datasets can be evaluated fast and in a uniform way. I have built a simulated database for the quantification method.

The challenge is to improve clinical routine evaluation based on MR spectroscopic data. Classification results will support diagnostic decision making.


Please sign in to download the files of this thesis.