Direct detection of neuronal activity through MRI is an active area of research [1]. It generally relies on the fact that ionic currents associated with synaptic activity result in small magnetic-field changes in the active brain area. These fields alter the precession frequency of the nuclear spins, leading to magnitude or phase changes. Typical MRI-based approaches for direct detection of neuronal currents use echo-planar imaging (EPI), which offers wide spatial coverage at low temporal resolution (>100ms). While the low spectral bandwidth of the acquired signal (<10Hz) is appropriate for detecting the delta, theta and perhaps alpha brain waves, it is not useful for detecting either the beta/gamma waves, or evoked potentials. Moreover, confounders of neuronal activation, such as vibrations caused by cold-head pumps, and subject cardiac- and respiratory-related motion, also have their signatures in the 0-3Hz domain [2]. These effects may increase the minimum detectable field changes to levels higher than could be caused by neuronal excitation.

In this work, we explore a significantly different paradigm, in which we give up wide spatial coverage, gaining the capability of sampling signals at rates significantly higher than 10Hz. Our simulation results indicate that the ability to compare entire time curves sampled at a high temporal rate (as opposed to single TE’s on those curves) leads to an increase in the sensitivity of detecting the effects of neuronal currents by a factor of at least 10.

Experiments were performed on a 3T, MR750 GE scanner, using 8/32 channel head coils. The water signal from a 64mm3 voxel in a 1g/l CuSO4-doped agarose gel phantom was sampled using a spectroscopic PRESS acquisition (Figure 1). 512 free induction decays (FID’s), sampled every 200ms, were collected using TR=1second (8.33min total acquisition time). Alternate FIDs were then artificially endowed with an additional phase change, corresponding to a small magnetic field step-function change (Figure 2). This could be a simulation of signal changes in the visual cortex, with the stimulus synchronized with the scanner’s every second acquisition. The average phase @TE=40ms and the average phase slope in the two signal pools (“excitation” ON and OFF) were then statistically compared for various excitation (ΔB₀) levels. This TE is typical for in vivo experiments [3], chosen as a compromise between maximizing contrast and minimizing SNR loss due to T2 decay. The minimum detectable ΔB₀ change was also assessed both for a single TE (the only option available for EPI imaging), and using the phase slope (if sampling the signal spectroscopically).Figure 3 presents an example of two FID phase signals, one acquired “on resonance” (blue dots), and one endowed with an additional neuronal current-induced phase offset of ΔB₀=0.2Hz (green dots); the off-resonance was exaggerated here for better visualization purposes. The time-dependent phase evolution is due to slight off-resonance behavior (<1Hz)—even in the voxel experimentally placed on resonance. The evolution under an additional off-resonance field causes a slope change. Note that while signal comparison at one given TE is noisy and requires signal averaging, slope comparison results in an immediate determination (in this case) of the difference between the two signals. Figure 4 presents a summary of our simulation results. Note that using our 256 signals per condition and the 32-channel coil, field changes as small as 0.001Hz (25 pT) can be detected if comparing phase slope; this is 20-fold higher than if using single TE acquisitions. This improvement reduces to a factor of 10, with the 8-channel coil. The higher detection capability of the 32-channel coil is expected, given the SNR increase/ phase-variability decrease achievable in this case. Last, but not least, note that while our approach requires trading off brain coverage for sensitivity increases, at least 3 voxels (suggested by standard fMRI or tractography, eg), can be independently quarried simultaneously. Any three independently chosen points define a plane, which can be excited by the first 90-degree pulse (Figure 1). Replacing the 2nd and 3rd refocusing pulses with multi-band pulses can result in simultaneous sampling of data from these three regions (Figure 5). Signal separation using Hadamard encoding or SENSE reconstruction can then be achieved.A simulation study is presented that suggests that trading off spatial coverage for temporal resolution can result in sensitivity increases of a factor of 20 to detect magnetic field changes caused by neuronal currents. Validation of these results in vivo is pending.