Abstract

In vivo 1H magnetic resonance spectroscopy (MRS) enables a non-invasive determination of tissue concentrations of various metabolites and compounds in animals or humans. Thus, metabolism can be investigated in vivo, and pathological as well as drug- or exercise-induced changes of the metabolism can be observed. MRS is increasingly being recognized as a valuable tool for the assessment of those anatomical regions - and related diseases - that are normally inaccessible for clinical examination (e.g. brain tumours).

In the past decade 3T MRI started to be routinely used for clinical examinations because, compared to the first 1.5T MRI spectrometers, it results in enhanced SNR and faster acquisitions. Nowadays, high field MRI spectrometers (3 to 7T and above) are used in many centres, mostly for research purposes. Higher magnetic field lead to better spatial resolution, while also increasing the separation of chemically distinct peaks, making it easier to attempt spectra profiling for metabolite identification. Despite this increase in spectral resolution, investigators and physicians hardly ever analyze the whole fingerprint of the spectrum which limits peak identification and quantification.

In this work we investigate the possibility to perform chemometric analysis on MRS datasets. To do so we have collected and analyzed localized spectroscopy, performed on patients diagnosed with various grades of glioma. Gliomas account for about 40% of total primitive brain tumours, and a discrimination between high and low glioma grades remains a vital diagnostic decision, determining the most effective treatment and having an important impact on the patient management and its outcome. We discuss the application of the Global Spectrum Deconvolution algorithm (GSD) on MRS data, followed by the generation of synthetic spectra to be used to generate the matrix used in subsequent statistical evaluation. GSD, made available in the Mnova software package of Mestrelab, is capable of identifying even poorly resolved spectral peaks and of fitting all recognizable peaks even in a very complex 1D spectrum in a surprisingly short time (typically a dozen seconds for up to 1000 peaks). It is fully automatic and objective (no human intervention is required) and produces a table of all detectable spectral peaks and their parameters. Such a table can be then used for various purposes like generation of artefact-free synthetic spectra (with or without resolution enhancement), stick spectra, artefact-free integrals, as well as accurate binning void of any bin-crossover problems due to the overlapping wings of spectral peaks.

Moreover, with the aim to identify metabolite peaks in an MRS spectrum using freely available metabolite databases (e.g.HMDB), we introduce the spin system definition for each standard metabolite. This allows us to take advantage of the spin simulation tool available under the used software to simulate the spin system at any desired magnetic field and nearly spin system size independently. The resulting simulated spectra are then employed to assign selected peaks in the MRS spectrum of interest.

This work will lead to the introduction of chemometric approaches to MRS data and to metabolite identification based on repositories acquired at different magnetic fields. In particular, for the dataset presented here these approaches can be integrated with histopathology and routine neuroimaging to facilitate the distinction between tumour types and grades.

Please, cite this online document as:
Mari S., Righi V., Nocetti L., Valentini A., Schenetti L., Marchioro C., Sykora S., Cobas C.,
   Toward direct application of multivariate statistical analysis to MRS Datasets,
   Poster at 54th ENC, Asilomar (CA), April 14-19, 2013, DOI: 10.3247/SL4Nmr13.001.