Korva is a scroll of the internet’s best ideas.
What MRI measures
MRI measures blood flow, so it sees how much energy different parts of the brain use over seconds. Each tiny unit of the image, called a voxel, often appears to have one uniform job. Language moves much faster than that, and a voxel can hide a lot.
Recording with Neuropixels
Neuropixels probes are long electrodes with many recording sites, and they can record from the same cortical column at many depths at once. Laura Williams's group was among the first to use them in this work. They pick up spiking, which is as close as possible to the actual timing of neural activity, so the blur of MRI goes away and responses can be time-locked to properties of the speech.
Tuning inside a voxel
Even within the depth of one cortical column, the recordings show rich differences in tuning. Some cells prefer certain speech properties and other cells prefer different ones. A voxel that looks uniformly tuned may be built from pieces with very different tuning, which raises the question of how circuit function can produce network function that looks so different, and how the network picture limits what is possible at the circuit level.
Why is fMRI so slow for language?
The MRI signal tracks blood oxygen, which changes only after neurons have been active for a moment. This blood-flow response peaks roughly four to six seconds after the neural activity that triggered it. Speech moves at a few words per second, so many words fit inside a single blood-flow response. The image therefore blends together events that the brain handled one after another.
How does a Neuropixels probe record so much?
A Neuropixels 1.0 probe has 960 recording sites along a shaft about one centimeter long, and 384 of them can be recorded at once, according to the 2017 Nature paper that introduced the design. The sites sit close together along the shaft, so a single probe samples many depths of cortex at the same time. The recordings are fast enough to resolve individual spikes. Versions adapted for use in human patients during surgery have since been reported.
Why is spiking closer to real neural timing?
A spike is an action potential, the electrical pulse a neuron sends to other neurons. Each one lasts about a millisecond, which is thousands of times faster than the blood-flow response MRI follows. Recording spikes therefore shows when a cell fired relative to a sound or a word. MRI shows only a delayed, summed echo of that activity.
More from Wu Tsai Neurosciences Institute, Stanford
