Discrimination of glioma brain tumor grades through Multivariate Data Analysis on 1H-HR-MAS NMR ex-vivo spectra

Gliomas account for about 40% of total primitive brain tumors, and discrimination between high and low glioma grades is a vital diagnostic decision, determining the most effective treatment and having an important impact on patient management and outcome. In vivo MRS can support the diagnosis of cancer based on MRI, but it can only be used when the molecular markers are well established. Their identification can be derived from the spectroscopic analysis of ex vivo biopsy samples using HR-MAS NMR technique. 45 specimens of brain tissue were obtained from 35 patients diagnosed either with glioma or lymphoma, and analyzed using HR-MAS NMR. With the aim of retrieving as much information as possible, three different pulse sequences were used, giving rise to three spectral datasets. Multivariate data analysis was performed to identify informative metabolites and their interactions. The key goals were: 1) explore the data; 2) integrate the spectra to enhance information retrieval and interpretability; 3) distinguish among glioma grades III and IV. After proper alignment [1], the datasets were explored by Principal Component Analysis (PCA). The spectra were integrated using Multivariate Curve Resolution (MCR, [2]) on selected intervals. MCR is a method able to resolve overlapping peaks and separate them, by extracting their “pure” peak shapes and their relative concentration in each sample. The concentration information was used to produce the interval-resolved data, which are in principle a more compact and cleaner version of the original data [3]. Class-modeling (SIMCA) and Discriminant Analysis (PLS-DA) were finally applied to the interval-resolved data considering each sequence dataset both separately and datafused. The differences in performance were assessed.