De-noising the MRS data is a key processing in analysis of spectroscopy MRS data. This paper presents an effective method based on wavelettransform and pattern recognition technologies. Upon the characteristics of MRS data, a new wavelet basis function was designed, and a de-noising method of Free Induction Decay (FID) data using wavelet threshold to obtain better MRS spectrums was conduced; hence, the features of some cancers from MRS spectrums based on Independent Component Analysis (ICA) and Support Vector Machine (SVM) were extended. Comparing with the de-nosing effect using conventional wavelet basis functions, experiments were conducted to validate that the innovative feature extraction method employing ICA and a new wavelet filter set has higher and better performance. Experiments in this study were carried out on a small amount of real and low SNR dataset that obtained from the GE NMR device. The experimental results showed that the proposed denosing method improves its efficiency of feature extraction significantly.