Optimised paediatric brain tumour diagnosis by using in vivo MRS and machine learning

Previous studies have shown that magnetic resonance spectroscopy (MRS) can provide diagnostic classifiers for childhood brain tumors. Here we investigate the effect of noise suppression on classification and build it into a pipeline aimed at optimizing the diagnosis. Although people have proposed several algorithms to suppress noise in MRS, the clinical application is still largely restricted to apodisation. We propose a wavelet-based framework to suppress the noise of in vivo MRS, thereby the signal quality of in vivo MRS and the classification accuracy of tumor cases was improved. The framework could be used in the clinical diagnosis through MRS.