Mapping BCI Modalities
Each of our techniques to read the brain is limited in some way. If we record noninvasively from the entire brain, we sacrifice spatial resolution. In order to resolve single neurons and make recordings at fine spatial resolution, we must insert a fine bundle of electrodes into the brain directly, which requires surgery. No single imaging modality has been non-invasive, high spatial resolution, and high temporal resolution simultaneously. For the past decades, neuroscientists have been working to balance these constraints.
I wanted to actually see that trade-off, so I put the main brain-computer interface modalities into a cube. Each axis is one of the things you are forced to trade:
Here are some takeaways from that map you can get by spinning it around in your mind. EEG and MEG are locked in a corner at very high temporal but very low spatial resolution, whereas fMRI and fNIRS are locked in another corner at very high spatial but very low temporal resolution (since they're measuring blood flow, which is very delayed compared to the underlying neural activity).
Now some of the more interesting nodes. ECoG manages to be high-high, since it just puts a grid on the cortex, but this comes at the cost of craniotomy. A really interesting new approach is the Stentrode, where an electrode-laden stent is inserted via a blood vessel to access motor cortex. No craniotomy, but with some cost in resolution. Finally, functional ultrasound is a new technology that lives somewhere in the middle of the cube - you gain very high (near cell level) spatial resolution by opening up a cranial window, but since it also measures blood flow, the temporal is limited.
Finally the last corner. Intracortical microwire arrays (whether the Utah array or the Neuralink approach) are by themselves with very high spatial and temporal but with very high invasiveness. So much of the work in the space can be thought of as trying to push this corner back toward the safe wall, by either having more invasive implants or by pushing the resolution of non-invasive techniques higher.
| Modality | Access | Spatial | Temporal |
|---|---|---|---|
| EEG | None (scalp) | ~cm | ~1 ms |
| MEG | None (helmet) | ~cm | ~1 ms |
| fNIRS | None (scalp) | ~cm | ~1 s |
| fMRI | None (scanner) | ~1 mm | ~1-2 s |
| Stentrode | Endovascular | ~few mm | ~1 ms |
| fUS | Cranial window | ~100 µm | ~100 ms |
| ECoG | Craniotomy (surface) | ~1 mm | ~1 ms |
| Intracortical | Penetrating implant | Single neuron | < 1 ms |
The axis values in the map are approximate and meant for shape, not precision. Real numbers vary a lot with electrode count, field strength, and exactly what you are trying to decode.