A compressive sensing spectral model for fNIRS haemodynamic response de-noising

In the biomedical scenario, near-infrared spectroscopy (NIRS) is employed as a non-invasive brain imaging technique. In particular, functional near-infrared spectroscopy (fNIRS) measures the brain response, also known as haemodynamic response (HR), to pre-defined stimuli. Processing of fNIRS data requires a great effort to extrapolate the informative component from a noisy mixture of physiological and spurious contributions. In this paper a novel fNIRS de-noising algorithm is presented and validated over both synthetic ideal and synthetic realistic data. The short-separation channel signal is divided into nonoverlapping short sequences. For each of them, a specific noise model is identified and subtracted from the corresponding standard channel data. The algorithm relies on a combination of a super-resolution technique based on Compressive Sensing theory and spectral analysis performed via Taylor-Fourier transform. Preliminary experimental results show a significant reduction of spurious components in all the considered conditions. No significant distortions are introduced in the recovered HR, ensuring reliable clinical interpretation of the acquired trace.