MRS data deconvolution through KBDM with multiple signal truncation and clustering: Circumventing noise effects

Abstract

This work describes an unsupervised Lorentzian model function-based deconvolution method for MRS applications through Krylov Basis Diagionalization Method (KBDM) using a strategy to circumvent noise effects sensibility of the direct application of the original algorithm. We simulate a brain MRS signal with known peaks parameters (amplitude, phase, frequency and transversal relaxation time) with SNR similar to the obtained in clinical settings and applied our method. The resulting peak list and the residual between the original signal and the estimated revealed the potential of the proposed method using multiple signal truncation in combination with clustering strategies. Further studies are needed in order to extend the applicability of this promising method, which will be crucial to demonstrate the potential value of the KBDM as a clinical MRS data processing tool.