What most people say is that fMRI correlates well with the local field potential (LFP), which is the EEG-like signal you get when an electrode records from neural tissue.The LFP, in turn, seems to be mostly correlated with “dendritic activity”, that is, incoming spikes to a brain area, not spikes generated by a brain area. Although the distinction between these two is vague, since local interneurons both receive and generate spikes.
The main problem with fMRI is that each 3D “voxel” [volumetric pixel] averages the activity of 100,000+ neurons over one second, so around 1 million spikes are being collapsed into a single average. The signal also lags by one second, and the signal is very faint. In fact the signal is so faint that many experimental trials need to be averaged, and then tiny task-dependent 1% differences in “activity” are thresholded to create colorful brain-like plots.
fMRI measures such a small and slow deviation from average activity levels, that it is a miracle that it correlates as well as it does with what is known about brain area specialization. And this is perhaps what makes it so compelling. Tasks that are known to involve a specialized brain area consistently show fMRI “activation” in those areas.
Still, while the colorful fMRI brain scan pictures are cool, it isn’t always clear what is being learned from them.
Want to learn more? Check out neuroscience grad student Shahab Bakhtiari’s answer below.
The exact Relationship between measured fMRI signal and neuronal activity is not completely understood yet, but there are some studies in fMRI literature that have tried to shed light on some dark side of BOLD signal whether by means of multimodal functional imaging (using electrophysiology, e.g., LFP, or multi-electrode spikes), or by recently established method Optogenetics (, )
In , the authors have investigated the correlation between
BOLD signal, single- and multi-electrode spike arrays, and Local field
To better understand the neural mechanisms underlying the BOLD response, spiking and synaptic activity were examined separately by analysing single- and multi-unit activity and LFPs, respectively.
The average LFP response was also found to give better estimates of the BOLD signal (smaller error in the least-square sense) when neural activity was convolved with the neural-vascular impulse response function than whenMUAwas used as the system’s input.
They also mention that:
The greater contribution of LFP activity to the fMRI signal is consistent with findings regarding the bioenergetics underlying this signal. It is well established that neural activity and energy metabolism are tightly coupled.
As these statements in discussion part of  mention, it seems that BOLD signal is mostly correlated with (not caused by) synaptic activities, rather than neuronal outputs (neurons’ spikes).
On the other hand, recent researches in this domain , thanks to the invention of optogenetics, declare that activation of local excitatory neurons gives rise to positive BOLD signal.
Using a novel integrated technology unifying optogenetic control of inputs with high-field fMRI signal readouts, we show here that specific stimulation of local CaMKIIa-expressing excitatory neurons, either in the neocortex or thalamus, elicits positive BOLD signals at the stimulus location with classical kinetics.
This research is well supported from a causality aspect, and we are less concerned regarding statistical spurious conclusions. Anyway, it seems that today we can be more assured than past times in using fMRI for inferences about neuronal activities.
References: K. Deisseroth, “Controlling the Brain with Light,” http://www.stanford.edu/g