Computational NeuroscienceReal-time automated spectral assessment of the BOLD response for neurofeedback at 3 and 7Â T

•Real-time single-voxel proton spectroscopy can assess T2* changes online.•Real-time single-voxel proton spectroscopy is feasible at 3 T and at 7 T.•FID optimized linear regression showed the highest sensitivity to BOLD.•Water peak can be approximated with a single complex lorentzian line in real-time.

Echo-planar imaging is the dominant functional MRI data acquisition scheme for evaluating the BOLD signal. To date, it remains the only approach providing neurofeedback from spatially localized brain activity. Real-time functional single-voxel proton spectroscopy (fSVPS) may be an alternative for spatially specific BOLD neurofeedback at 7 T because it allows for a precise estimation of the local T2* signal, EPI-specific artifacts may be avoided, and the signal contrast may increase. In order to explore and optimize this alternative neurofeedback approach, we tested fully automated real-time fSVPS spectral estimation procedures to approximate T2* BOLD signal changes from the unsuppressed water peak, i.e. lorentzian non-linear complex spectral fit (LNLCSF) in frequency and frequency-time domain. The proposed approaches do not require additional spectroscopic localizers in contrast to conventional T2* approximation based on linear regression of the free induction decay (FID). For methods comparison, we evaluated quality measures for signals from the motor and the visual cortex as well as a real-time feedback condition at high (3 T) and at ultra-high (7 T) magnetic field strengths. Using these methods, we achieved reliable and fast water peak spectral parameter estimations. At 7 T, we observed an absolute increase of spectra line narrowing due to the BOLD effect, but quality measures did not improve due to artifactual line broadening. Overall, the automated fSVPS approach can be used to assess dynamic spectral changes in real-time, and to provide localized T2* neurofeedback at 3 and 7 T.