Spectral denoising for MR spectroscopy using noise level function

Abstract

1H-MRSI (proton Magnetic Resonance Spectroscopic Imaging) is now widely used to assist physicists to analyze and quantify brain metabolites in a noninvasive way. In case of glioma, the brain tissue metabolite composition is not widely different but metabolites concentration are varying depending on the grade of the tumor. In the higher stage of the tumor, new metabolites could be detected such as lactate or lipids in the long and short echo time. These variations appear owing to brain tissue modification generated by the tumoral process and help to classify the tumor in grades. 1H-MRS provides crucial data to feed models in order to estimate the best treatment (surgical, etc). Initially, the spectrum is the result of a large number of acquisitions. With a sufficiently long acquisition time, the resulting signal resulting from about 150 mean filter appears to be very low in noise. Nevertheless in this case acquisition time is too long to envisage the study of different parts of the brain. That is why we need to propose an efficient and robust spectrum denoising method.