Spectral Wavelet-feature Analysis and Classification Assisted Denoising for enhancing magnetic resonance spectroscopy.

Magnetic resonance spectroscopy (MRS) is capable of revealing important biochemical and metabolic information of tissues noninvasively. However, the low concentrations of metabolites often lead to poor signal-to-noise ratio (SNR) and a long acquisition time. Therefore, the applications of MRS in detection and quantitative measurements of metabolites in vivo remain limited. Reducing or even eliminating noise can improve SNR sufficiently to obtain high quality spectra in addition to increasing the number of signal averaging (NSA) or the field strength, both of which are limited in clinical applications. We present a Spectral Wavelet-feature ANalysis and Classification Assisted Denoising (SWANCAD) approach to differentiate signal and noise peaks in magnetic resonance spectra based on their respective wavelet features, followed by removing the identified noise components to improve SNR. The performance of this new denoising approach was evaluated by measuring and comparing SNRs and quantified metabolite levels of low NSA spectra (e.g. NSA = 8) before and after denoising using the SWANCAD approach or by conventional spectral fitting and denoising methods, such as LCModel and wavelet threshold methods, as well as the high NSA spectra (e.g. NSA = 192) recorded in the same sampling volumes. The results demonstrated that SWANCAD offers a more effective way to detect the signals and improve SNR by removing noise from the noisy spectra collected with low NSA or in the subminute scan time (e.g. NSA = 8 or 16 s). The potential applications of SWANCAD include using low NSA to accelerate MRS acquisition while maintaining adequate spectroscopic information for detection and quantification of the metabolites of interest when a limited time is available for an MRS examination in the clinical setting.