Feature analysis and denoising of MRS data based on pattern recognition and wavelet transform

Abstract

Denoising the MRS data to provide better data sources and feature analysis of spectroscopy are the main concerns in MRS data processing. This paper describes an effective method based on wavelet transformation and pattern recognition technologies. According to the characteristics of MRS data, a new wavelet base function was designed, and denoising of FID data was performed by using wavelet threshold to obtain better MRS spectra firstly, then extracted the feature of certain cancers from MRS spectra based on independent component analysis (ICA) and support vector machine (SVM). Contrast with the denoising effect of conventional wavelet base functions, the experimental results confirmed the validity of the feature extraction method of ICA, and the newly-designed wavelet filter set showed better performance. Experiments were carried out on small amounts of very low SNR datasets which were obtained from the GE NMR device, and the results showed the improved effect on denoising and feature extraction.