Smooth dynamic T 2 * mapping in fMRI based on a novel, total variation-minimizing algorithm for efficient multi-echo BOLD-time series denoising with high signal-to-noise (SNR) and contrast-to-noise (CNR) ratios

Jan Michálek^1*Michal Mikl²

• ¹Faculty of Informatics, Masaryk University, Brno, Czechia
• ²Central European Institute of Technology (CEITEC), Brno, Olomouc, Czechia

This report deals with advanced processing of blood oxygenation-dependent (BOLD) functional magnetic resonance (fMRI) signals. It does not address functional characteristics of the human cortex such as functional connectivity. fMRI is based on measurement of BOLD variations of transverse relaxation time T2* or T2 . T2*or T2 can be calculated when multiple echoes of the MRI signal are recorded and may be more resistant to artifacts or better characterize tissue properties than the echoes themselves.To develop a robust-to-noise algorithm for dynamic T2 * mapping from a three gradient echo (GRE) signal, allowing to explore the potential of quantitative T2* mapping.METHODS: fMRI resting-state and block-design visual task three-echo data were acquired from nine healthy volunteers. A significant problem in multi-echo T2* fitting is the noise in the echoes. The majority of BOLD-denoising methods first pinpoint some source of noise and subsequently remove the respective noise time series. We instead first postulated that the blood oxygenation changes smoothly, and consequently developed a state-of-the-art, total variation-minimizing, denoising algorithm that simultaneously enforces smoothness of the processed BOLD echoes and preserves local temporal signal means. To make calculated T2* time courses also smooth, they were estimated from total variation (TV) denoised echoes. We used a denoising approach initially proposed by Professor Stanley Osher for two-dimensional (2D) images that has been very successful, most prominently in space research, where it enabled the reconstruction of the first ever image of a black hole. To our knowledge, Osher's approach has so far not been used elsewhere for denoising of one-dimensional fMRI time series.Signal-to-noise and contrast-to-noise distributions of the denoised echoes, as well as of the T 2 * time series, were superior to those obtained by the current fMRI denoising methods (3dDespike, tedana, NORDIC). The denoised echoes and the T2* time courses match the shape of the theoretical hemodynamic function much better than previous results.The total-variation-minimizing fMRI time series denoising algorithm yields denoised echoes of unprecedented quality that allow estimation of smooth dynamic T2* maps, i.e. transition from qualitative-only fMRI echoes to fMRI signal endowed with time units.