Community /

The Building Blocks of the Human Brain – Part II


This is a 2-part series of articles. The first one is a tour of brain anatomy whilst the second is about two major non-invasive approaches we can use to record brain activity. Click here for part 1.


Most of the structures we’ve covered unfortunately lie too deep to study without surgical implantation of electrodes. Fortunately, the superficial location of the Cerebral Cortex makes it amenable to study by bio-hackers and researchers using affordable yet powerful non-invasive techniques such as EEG and fNIRS.

EEG

Mechanism 

In EEG (electroencephalogram), we place electrodes on the scalp to directly sense the electrical fields generated by the synaptic activity of neurons inside the brain. Because the neurons of the cerebral cortex are so well lined up geometrically, their small individual electrical fields add up to fields large enough that we can pick them up from the scalp.

Advantages of EEG

EEG has several advantages going for it:

  1. Excellent signal processing pipelines. Being one of the most widely-used techniques in neuroscience and part of the larger family of electrophysiology, there is an abundance of literature and software packages for analysing, processing and de-noising EEG signals.
  2. Speed and responsiveness. EEG is sensitive to sub-millisecond fluctuations in brain activity. Your limitation is only the sampling rate of the equipment you’re using to record.
  3. Direct measure. EEG reads the exact modality which neurons use to communicate: electrical currents. There is no proxy, correlate or transformation involved, what you see is the same “language” that the brain is speaking.
  4. Well understood. We have excellent knowledge of the biophysics underlying brain activity from the level of synapses to massive populations of neurons. We also have almost 100 years of history relating EEG waveforms to specific processes in perception, cognition and action, which helps us interpret the signal.

Below, we can see what EEG data looks like, in a screengrab from the OpenBCI GUI. Each trace from top to bottom indicates the electrical activity detected from an electrode on a different position on the scalp. The top purple trace shows how an eye blink appears. The trace corresponds not to the neural processes that control the eye blink, but instead to the electrical field that is generated during its occurrence. The regions flanking either side of the eye blink show normal EEG data. 

The red trace has highlighted a period containing strong Alpha activity. Brain activity recorded using EEG is traditionally subdivided into frequency bands, which correspond to the different rhythms that neurons and cortical regions use to communicate.

Alpha is the name given to the frequency band of 8-12 Hz and typically occurs when subjects are relaxed with their eyes closed but awake. Gamma (20-40 Hz) is another frequency band of interest, which usually occurs when a cortical region is highly engaged during an attentive process. On the right side (top) we can see a Fast Fourier Transform (FFT) plot. The FFT is an algorithm for decomposing a complex signal into its component frequencies and plotting the amplitude or power of each. This allows a user to easily get an idea of the dynamics of cortical activity, that is how active and in what kind of activity each area inspected is currently.

Apart from studying the frequencies of ongoing brain activity, we can also look at the waveforms themselves, in response to different types of stimuli. These are called Evoked Potentials and have been very well characterised by almost 100 years of research. We have a strong understanding of Evoked Potentials generated by presented sensory stimuli, as well as some cognitive processes such as mistakes (Error-Related Negativity) and unexpected stimuli (P300 and “oddball” paradigms).

Disadvantages of EEG

Although the signals recorded with EEG and invasive electrophysiology share a mechanistic basis, the presence of the skull limits our readings, compared to what we would get with implanted electrodes, in the following ways:

  1. We can only see a spatial average of the activity of 10M-100M neurons. The electrical fields generated by a single neuron are too small to be detected more than 0.1 millimetres away from it. However, the fact that most neurons in Cortex are arranged stereotypically in parallel means that their individual electrical fields add up. If enough neurons are active at more or less the same time, the electrical field in that patch of brain tissue sums constructively and becomes large enough to be detectable by overlying electrodes on the scalp.
  2. We can only see slow electrical signals, not action potentials. Action potentials peak and return to baseline within 2-5 ms and it is extremely unlikely that enough (10k-100k) neurons spike at the exact same time for their signals to add up constructively and be detectable. As such, the signals we detect with EEG mainly correspond to slow synaptic potentials, synchronously happening across thousands of neurons at under 100 Hz.
  3. We can confidently locate signals to about 10 mm. Brain tissue is a relatively good insulator, meaning electrical currents don’t spread too far from their sources. However, underneath the skull is a sheet of Cerebrospinal Fluid, an electrolyte-filled liquid which is an excellent conductor and can therefore lead to “smearing” of electrical currents, adding uncertainty to the cortical area they originate from.
  4. The surface location of EEG makes it vulnerable to certain contaminants. Brain activity is not the only electrical signal on the head. The eyes and head muscles generate powerful electrical fields which are also picked up by EEG, since electrodes are evidently agnostic to the source of currents they detect. These eye and face signals can and should be exploited, but when analysing brain data one must take care to filter them out when making inferences about brain and behaviour. 

fNIRS

Mechanism

Functional near-infrared spectroscopy (fNIRS) measures brain activity indirectly: instead of sensing the electrical signals that neurons produce when they’re active, it tracks how much “fuel” they’re burning by looking at changes in local blood oxygen levels. It achieves this by pumping light of 2 different wavelengths (typically in the green and red bands) into the brain, through the skull, and measuring, with a photodetector placed on the scalp nearby, either how much light is absorbed (Continuous Wave-fNIRS), how much light is delayed (Temporal Difference-fNIRS) or how perturbed its phase is (Frequency Domain-fNIRS) by the portion of brain tissue it travels through before it reaches the photodetector. Because oxygenated and de-oxygenated blood absorb greenish and reddish light differently, by comparing between their levels it is possible to infer how oxygenated the blood in the examined region of the brain is, and estimate how active it is.

Advantages of fNIRS

  1. Large signal to noise ratio. Changes in blood flow and blood oxygenation levels integrate over several seconds into massive signals. With appropriate controls and signal processing pipelines, this gives fNIRS and enviable signal to noise ratio.
  2. Potentially better resolution. Sophisticated forms of fNIRS (high definition diffusive optical tomography) can achieve spatial resolutions between 5 and 10 millimetres by employing massive arrays of light sources and detectors, special arrangements in their positioning and sophisticated analysis algorithms for estimating where a bout of brain activity originated from.

Disadvantages of fNIRS

As a non-invasive optical technique, fNIRS faces a few challenges:

  1. It measures an indirect correlate of brain activity. Blood oxygen consumption is not a direct index of brain activity and it is still not clear how changes in blood oxygen actually map onto brain activity. 
  2. It is very slow. Changes in blood oxygen levels happen extremely slowly, at the seconds time-scale, whereas we know neural synaptic and action potential signals happen in the millisecond time-scale. This immediately rules out fNIRS for looking at fast sensory, motor and cognitive processes or BCI applications which require fast responsiveness.
  3. It is also subject to contaminations. Blood oxygen changes can happen anywhere in head tissues, not just the brain, and will show on fNIRS readings. Changes in systemic blood supply or oxygenation can also introduce biases and errors into fNIRS measurements. 
  4. Photons don’t penetrate the brain very well. Unsurprisingly, the majority of photons pumped through the skull by fNIRS never find their way out. This limits fNIRS to imaging the superficial layers of the cortex (top 1-2 millimetres), resulting in a potentially biased view of cortical activity because of the different functions of each cortical layer.

Conclusion

There are no free lunches in neurotechnology. Every technique has advantages and drawbacks and it’s hard to say if one is better than the other. This is certainly the case for EEG and fNIRS, which excel at different metrics. EEG has unsurpassed temporal resolution and mechanistic understanding, whereas fNIRS can achieve impressive spatial resolution for a headset and measures strong (if highly indirect) signals. Which one to use will depend on the scientific question being asked or the application or product developed.

Though both technologies have a long history – EEG is almost 100 years old and fNIRS just turned 40 – they are far from obsolete. The last years have seen exciting innovations and I expect there is a lot more to come from these two players.

About the Author

André Marques-Smith is a Scottish-Portuguese Neuroscientist, with a PhD from Oxford University and a postdoctoral career spanning King’s College London and the Sainsbury-Wellcome Centre for Neural Circuits and Behaviour, where he held a Henry-Wellcome fellowship. After early research in human neuropsychology, André worked on mapping developing cortical circuits in mice using electrophysiology and optogenetics. He went on to work on neurotechnology (electrocorticogram and Neuropixels invasive probes) and Systems Neuroscience (connections of the thalamus and behaviour), before joining a venture-backed non-invasive BCI startup in London. He is now CTO and co-founder of a new startup applying state-of-the-art neuroscience outside the lab.

Find me on LinkedIn or Twitter and check out my Bookshlf on Neurotech & BCI.

One Comment

Leave a Reply